process high data rate stream system proceed confer intellig data understand agent data mine interact central featur framework avoid ontolog agre meta languag format support system wrapper advantag offer system easili extend data agent mine agent simpli ad system demonstr emad framework paper includ detail emad architectur wrapper principl incorpor full descript framework oper provid aedm scenario scenario focu evalu framework wavelet decomposit uncertain time seri data set orient distanc dbase discrimin featur extract multi dclass classif detect leader correl time seri mine divers network pocket data mine collabor data mine mobil comput environ dgrowth effici algorithm high util itemset mine dsupport anonym base pseudo taxonomi outsourc frequent itemset mine semi dsupervis featur select graph classif detect abnorm coupl sequenc sequenc group dbase manipul trade behavior basset scalabl gatewai finder larg graph onlin mine tempor maxim util itemset data stream cluster graph stream cluster graph stream predict model heterogen sourc mine knowledg databas inform network analysi approach mine concept ddrift data stream recent mine data stream concept drift action insight import challeng task wide rang applic includ credit card fraud protect target market network intrus detect convent knowledg discoveri tool face challeng overwhelm volum stream data concept drift paper propos gener framework mine concept drift data stream weight ensembl classifi train ensembl classif model ripper naiv bayesian sequenti chunk data stream classifi ensembl judici weight base expect classif accuraci test data time evolv environ ensembl approach improv effici learn model accuraci perform classif empir studi show propos method substanti advantag singl classifi approach predict accuraci ensembl framework effect varieti classif model privaci dpreserv data publish survei recent develop collect digit inform govern corpor individu creat tremend opportun knowledg inform base decis make driven mutual benefit regul requir data publish demand exchang public data parti data origin form typic sensit inform individu publish data violat individu privaci current practic data publish reli polici guidelin type data publish agreement publish data approach lead excess data distort insuffici protect privaci preserv data publish ppdp method tool publish inform preserv data privaci recent ppdp receiv consider attent research commun approach propos data publish scenario survei systemat summar evalu approach ppdp studi challeng practic data publish clarifi differ requir distinguish ppdp relat problem propos futur research direct music recommend content context inform mine guest editor introduct social comput blogospher introduct domain ddriven data mine special section cluster massiv text categor data stream applic new group filter text crawl document organ requir real time cluster segment text data record categor data stream cluster problem number applic problem custom segment real time trend analysi present onlin approach cluster massiv text categor data stream statist summar methodolog present result illustr effect techniqu gener framework relat graph cluster gener framework relat graph cluster dens pattern mine graph stream massiv web commun networkappl creat data repres massiv sequenti stream edg convers telecommun network messag social network repres massiv stream edg stream typic larg larg amount underli activ network import applic domain determin frequent occur dens structur underli graph stream gener determin frequent dens pattern underli interact introduc model dens pattern mine propos probabilist algorithm determin structur pattern effect effici purpos probabilist approach creat summar graph stream pattern mine show summar approach lead effect effici result stream pattern mine number real synthet data set person privaci protect social network discrimin frequent subgraph mine optim guarante framework cluster massiv graph stream optim distanc bound fast search compress time dseri queri log right protect trajectori dataset nearest dneighbor preserv icdm abstract assist domain expert activ learn select construct fewer exampl request label build accur classifi previou work activ learn gener specif queri real world applic domain expert oracl readili answer gener queri don care attribut power gener queri gener queri equival specif overli gener queri good answer domain expert oracl highli uncertain make learn difficult paper propos activ learn algorithm ask good gener queri extend algorithm construct hierarch featur nomin numer attribut demonstr experiment method ask significantli fewer queri compar previou work activ learn initi label dataset small oracl inaccur class probabl estim method readili deploi real world data mine task obtain label exampl costli extens paper publish ieee icdm keyword activ learn domain expert gener queri icdm abstract assist domain expert activ learn select construct fewer exampl request label build accur classifi previou work activ learn gener specif queri real world applic domain expert oracl readili answer gener queri don care attribut power gener queri gener queri equival specif overli gener queri good answer domain expert oracl highli uncertain make learn difficult paper propos activ learn algorithm ask good gener queri extend algorithm construct hierarch featur nomin numer attribut demonstr experiment method ask significantli fewer queri compar previou work activ learn initi label dataset small oracl inaccur class probabl estim method readili deploi real world data mine task obtain label exampl costli extens paper publish ieee icdm keyword activ learn domain expert gener queri agent data mine interact central featur framework avoid ontolog agre meta languag format support system wrapper advantag offer system easili extend data agent mine agent simpli ad system demonstr emad framework paper includ detail emad architectur wrapper principl incorpor full descript framework oper provid aedm scenario scenario focu evalu framework onlin auction scalabl olap mine inform network proud probabilist approach process similar queri uncertain data stream distribut process spatial alarm safe region dbase approach spatial alarm consid basic capabl futur mobil comput system enabl person locat base servic paper propos distribut architectur suit safe region techniqu scalabl process spatial alarm show safe region base process enabl resourc optim distribut partial alarm process task server mobil client propos safe region comput algorithm explor impact size shape safe region network bandwidth server load client energi consumpt concret show maximum weight perimet rectangular safe region approach outperform previou techniqu term perform accuraci explor finer granular safe region introduc grid base pyramidbas represent rectilinear polygon shape bitmap encod experiment evalu show distribut safe region base architectur outperform popular server centric approach period safe period base spatial alarm process privaci dpreserv singular decomposit scale dup strategi process high drate data stream system effici queri process stream count cell processor naiv ensembl work cloud comput effici increment mine qualifi web travers pattern scan origin databas earli predict time seri nearest neighbor approach paper formul problem earli classif time seri data import time sensit applic healthinformat introduc concept mpl minimum predict length develop ect earli classif time seri effect nearest neighbor classif method ect make earli predict time retain accuraci compar nn classifi full length time seri empir studi benchmark time seri data set show ect work real data set nn classif effect doptim supervis featur select frequent subgraph graph classif increasingli import step numer applic domain function predict molecul protein computeris scene analysi anomali detect program flow approach propos literatur graph classif base frequent subgraph popular branch graph repres binari vector compon indic graph subgraph frequent dataset larg graph face enorm problem number frequent subgraph grow exponenti size graph possess discrimin power make optim distanc bound time dseri data data mine oper includ integr search compon core perform similar search classif base nearest neighbor larg depend underli compress distanc estim techniqu data repositori grow larger explicit store data compress form facilit mine oper directli compress data natur qualiti tight estim distanc compress object directli affect search perform motiv work set search engin weblog repositori keyword demand trend time repres store compress timeseri data search analysi sequenc data import applic search engin includ discoveri import new event keyword recommend effici keyword advertis map present mechan fast search oper compress time seri data specif focu weblog data import contribut work deriv optim tight bound euclidean distanc estim compress sequenc methodolog applic sequenti data gener propos techniqu independ interest addition distanc estim strategi ti specif compress methodolog appli top orthonorm base compress techniqu fourier wavelet pca experiment result optim bound lead signific improv prune power search compar previou state art case elimin candid search sequenc posit unlabel learn data stream classif anonym worst dcase distribut dbase background knowledg survei uncertain data algorithm applic inform discoveri multipl stream paper address issu continu keyword queri multipl textual stream explor techniqu extract inform paper repres knowledg approach perform keyword search multipl textual stream scenario intuit assum research financi analyst search inform topic continu poll data multipl possibl heterogen text stream rss feed blog topic interest embed retriev privat metadata electrocardiogram answer linear optim queri approxim stream index graph olap multi ddimension framework graph data analysi dsesam web tool dterm dbase gene similar analysi knowledg discoveri effici method maxim bichromat revers nearest neighbor bichromat revers nearest neighbor brnn extens studi spatial databas literatur paper studi relat problem call maxbrnn find optim region maxim size brnn problem real life applic includ problem find server point attract custom proxim straightforward approach determin brnn point feasibl larg infinit number point knowledg fastest method exponenti time complex data size base interest properti problem effici algorithm call maxoverlap extens experi conduct show algorithm time faster techniqu gconnect connect index massiv disk dresid graph problem connect extrem import context massiv graph larg commun network social network graph desir determin minimum cut pair node problem solv classic literatur relat maximum flow problem effici solvabl larg graph disk resid graph effici process connect queri minimum cut problem typic solv varieti combinatori flow base techniqu requir random access underli edg graph paper propos develop connect index massiv disk resid graph edgesampl base approach creat compress represent underli graph compress represent held main memori deriv effici approxim minimum cut problem compress represent organ disk resid index structur present experiment result show result approach order magnitud effici queri process disk resid approach expens small amount accuraci introduct special issu social comput densiti dbase cluster data stream multipl resolut onlin pair voip convers cluster complex graph complex graph multi type node link frequent aris import applic web mine inform retriev bioinformat epidemiolog studi propos gener framework cluster complex graph framework deriv famili cluster algorithm includ hard soft version capabl learn cluster pattern complex graph structur statist properti establish connect propos framework tradit graph partit approach experiment evalu encourag result valid propos framework algorithm content dbase filter effici onlin materi view mainten real dtime event detect video stream dual encrypt queri integr assur hierarch paramet dfree commun discoveri abstract larg bipartit graph document term userproduct graph find meaning commun quickli automat propos commun hierarchi commun commun propos method context specif cluster tree cct find commun multipl level user intervent base inform theoret principl mdl specif partit graph progress refin subgraph allow user quickli navig global coars structur graph focus local pattern fring benefit addit indic qualiti achiev compress typic hierarch method demonstr scalabl effect real larg graph fast comput reachabl label larg graph high compress rate numer applic deal larg graph queri reachabl node graph hop cover compactli repres edg transit closur graph space answer reachabl queri effici challeng comput hop cover exist approach suffer larg resourc consumpt low compress rate paper propos hierarch partit approach partit larg graph subgraph repeatedli top fashion uniqu featur approach comput hop cover partit iter top partit provid techniqu comput hop cover connect subgraph cover comput cut graph subgraph result cover high compress entir graph approach propos node orient approach edg orient approach approach effici comput hop cover larg graph high compress rate extens experi studi show hop cover graph node billion connect obtain minut compress rate pc ownership protect shape dataset geodes distanc preserv scalabl process spatial alarm spatial alarm extend idea time base alarm spatial dimens time base alarm set remind arriv futur refer time point spatial alarm set spatial locat interest subscrib alarm travel futur spatial alarm process requir meet demand object high accuraci ensur low alarm miss high scalabl requir highli effici optim process spatial alarm paper present motion awar framework facilit systemat method scalabl process spatial alarm introduc concept safe period minim number unnecessari spatial alarm evalu increas throughput scalabl system show safe period base alarm evalu techniqu significantli reduc server load spatial alarm process compar period evalu approach preserv accuraci timeli spatial alarm develop suit spatial alarm group techniqu base spatial local alarm motion behavior mobil user reduc safe period comput cost spatial alarm evalu server side evalu scalabl accuraci approach road network simul show propos motion awar safe period base approach spatial alarm process offer signific perform enhanc alarm process server maintain high accuraci spatial alarm compar convent period alarm evalu approach predict failur manag distribut stream process system right protect trajectori dataset onlin failur forecast fault dtoler data stream process high dimension index uncertain data framework cluster uncertain data stream direct discrimin pattern mine effect classif classif effici construct compact shed filter data stream process locust onlin analyt process framework high dimension classif data stream sampl dbase approach inform recoveri abstract recent resurg interest research noisi incomplet data applic requir inform recov data ideal approach inform recoveri featur incorpor prior knowledg data knowledg form complex distribut constraint close form solut exist captur complex correl quantifi degre uncertainti recov data support queri data databas commun develop number approach inform recoveri gener offer featur overcom limit significantli gener approach inform recoveri base sampl appli sequenti import sampl techniqu statist work complex distribut dramat outperform naiv sampl data constrain illustr gener effici approach applic scenario cleans rfid data recov inform publish data summar random privaci fast graph pattern match stop chase trend discov high order model evolv data abstract applic driven evolv data pattern web traffic program execut trace network event log stationari build predict model evolv data import challeng task approach work chase trend learn updat model evolv data impromptu model onlin predict case prove costli ineffect time wast learn recur concept classifi remain step current trend time paper propos mine high order model evolv data limit number concept stabl distribut data stream concept switch constantli mine concept offlin histor stream build high qualiti model run time combin histor concept chang pattern cue provid onlin train stream find current concept model classifi data unlabel stream primari advantag high order model approach high accuraci experi show benchmark dataset classif error highord model small fraction current approach import benefit unlik state theart approach approach requir user tune paramet achiev satisfi result stream characterist graph olap onlin analyt process graph research interest gener interest data mine data wareh databas system current research focus model manag analyz larg scale graph inform network data dirichlet process base evolutionari cluster evolutionari cluster emerg import research topic recent literatur data mine solut problem found wide spectrum applic social network analysi paper base recent literatur dirichlet process develop specif model solut problem dpchain hdp evo model substanti advanc literatur evolutionari cluster sens perform exist literatur importantli capabl automat learn cluster number structur evolut extens evalu demonstr effect promis model state art literatur evolutionari cluster hierarch dirichlet process hidden markov state paper studi evolutionari cluster recent hot topic import applic notic social network analysi paper base recent literatur hierarch dirichlet process hdp hidden markov model hmm develop statist model hdp htm combin hdp hierarch transit matrix htm base propos infinit hierarch hidden markov state model ih ms effect solut problem hdp htm model substanti advanc literatur evolutionari cluster sens perform exist literatur importantli capabl automat learn cluster number structur time explicitli address correspond issu evolut extens evalu demonstr effect promis solut state art literatur time sensit rank applic public search multi dspace dmap svm multi dclass classif context ddriven content adapt planner improv mobil internet access direct mine discrimin essenti frequent pattern model dbase search tree anonym transact databas public paper consid problem publish transact data research purpos transact arbitrari set item chosen larg univers detail transact data electron imag life implic transact data excel candid data mine research transact data rais concern individu privaci transact data releas data mine made anonym data subject identifi challeng transact data structur extrem high dimension tradit anonym method lose inform data date satisfactori privaci notion solut propos anonym transact data paper propos address issu forward semi dsupervis featur select outlier detect uncertain data recent year techniqu develop mine manag uncertain data wai collect data result enorm amount inconsist miss data data remodel form uncertain data paper examin problem outlier detect uncertain data set outlier detect problem challeng uncertain case outlier behavior data point result uncertainti ad data point uncertainti ad data point skew data distribut true outlier mask critic remov effect uncertainti ad aggreg level level individu data point paper examin densiti base approach outlier detect show remov uncertainti underli data present experiment result illustr effect method type dindepend correct sampl select bia structur discoveri dbalanc sampl select bia common problem real world applic train data obtain realist constraint make follow distribut futur test data applic hospit clinic studi common practic build model elig volunt train data appli model entir popul volunt select random train set drawn distribut test set dataset suffer sampl select bia covari shift past year work propos reduc sampl select bia static match distribut train set test set paper explor distribut directli propos index high demension data uncertainti mine sequenc classifi earli predict supervis learn sequenc data sequenc classif recogn import data mine task signific applic tempor order import sequenc data critic applic sequenc classif medic diagnosi disast predict earli predict highli desir featur sequenc classifi earli predict sequenc classifi prefix sequenc short make accur predict knowledg earli predict sequenc data studi systemat paper identifi problem mine sequenc classifi earli predict analyz problem challeng attempt tackl problem propos interest method sequenti classif rule scr method mine set sequenti classif rule classifi call earli predict util defin select featur rule gener sequenti decis tree gsdt method adopt divid conquer strategi gener classif model conduct extens empir evalu real data set interestingli method achiev accuraci compar stateof art method typic short prefix sequenc result earli predict highli feasibl effect proxim track time devolv bipartit graph author confer network evolv time confer author close relat chang time larg time evolv bipartit graph set social network citat market basket analysi collabor filter goal monitor central individu node import author ii proxim node set node import author respect confer effici increment provid time answer propos ptrack ctrack base random walk restart power matrix tool experi real data show method effect effici mine result agre intuit achiev time speed qualiti loss gener model multipl view unsupervis learn multipl view data multipl represent featur space graph space aris data mine applic inform retriev bioinformat social network analysi represent statist properti learn consensu pattern multipl represent challeng problem paper propos gener model multipl view unsupervis learn propos model introduc concept map function make pattern pattern space compar optim pattern learn multipl pattern multipl represent model formul specif model spade system declar stream process engin paper present spade system declar stream process engin system larg scale distribut data stream process middlewar develop ibm watson research center front end rapid applic develop system spade intermedi languag flexibl composit parallel distribut data flow graph toolkit type gener built stream process oper support scalar vector process seamlessli inter oper user defin oper rich set stream adapt ingest publish data sourc importantli spade automat bring perform optim scalabl system applic end spade emploi code gener framework creat highli optim applic run nativ stream process core spc execut commun substrat system full advantag system servic spade develop construct applic fine granular stream oper worri perform implic exist distribut system spade optim compil automat map applic appropri size execut unit order minim commun overhead time exploit parallel virtu scalabl system runtim spade effect code gener optim scale applic larg number node run spade job processor physic node tightli connect cluster environ spade ibm research creat real world stream appli mine signific graph pattern leap search temperatur dawar schedul system dthrottl good abstract power awar oper system ensur system temperatur exce threshold util system throttl techniqu system load altern clock speed scale temperatur hit threshold time system oper maximum load paper show simpl system throttl rule fact achiev assumpt show maintain constant oper speed temperatur work oper altern period cool heat result set reason temperatur model prove system throttl effect temperatur awar schedul natur assumpt hold discuss scenario assumpt relax argu complex schedul algorithm case introduct energi temperatur manag processor system increasingli import problem power consumpt chip rise drastic gener time rate technolog improv cool system keep pace natur result larg bodi work attempt incorpor energi temperatur consider processor schedul level incorpor system level chip architectur scale speed processor hot advanc processor technolog oper system level modern dai processor interfac user control speed real time current processor intel amd ibm mechan call dynam voltag scale dv control clock speed processor vari suppli voltag oper system includ command user access interfac cpufreq linux identifi influenti blogger commun blog popular web user publish inform web blogger write blog post share like dislik voic opinion provid suggest report new form group blogospher blogger form virtual commun similar interest activ happen blogospher affect extern world understand develop blogospher find influenti blog site influenti blog site form long tail blog site influenti influenti blogger inspir high impact influenti physic commun studi problem identifi influenti blogger blog site activ blogger necessarili influenti influenti blogger impact fellow blogger wai paper discuss challeng identifi influenti blogger investig constitut influenti blogger present preliminari model attempt quantifi influenti blogger pave build robust model find type influenti illustr issu conduct experi data real world blog site evalu multi facet problem identifi influenti blogger discuss uniqu challeng conclud interest find futur work holist lexicon dbase approach opinion mine import type inform web opinion express user gener content custom review product forum post blog paper focu custom review product studi problem determin semant orient posit neg neutral opinion express product featur review problem applic opinion mine summar search exist techniqu util list opinion bear word call opinion lexicon purpos opinion word word express desir great amaz undesir bad poor state approach major shortcom paper propos holist lexicon base approach solv problem exploit extern evid linguist convent natur languag express approach system handl opinion word context depend major difficulti exist algorithm deal special word phrase languag construct impact opinion base linguist pattern effect function aggreg multipl conflict opinion word sentenc system call opinion observ base propos techniqu implement experiment result benchmark product review data set addit review show propos techniqu highli effect outperform exist method significantli approxim frequent itemset mine presenc random nois static dynam method condens dbase privaci dpreserv data mine recent year privaci preserv data mine import problem larg amount person data track busi applic case user unwil provid person inform privaci sensit inform guarante paper propos framework privaci preserv data mine multi dimension data previou work privaci preserv data mine perturb approach reconstruct data distribut order perform mine approach treat dimens independ ignor correl dimens addit requir develop distribut base algorithm data mine problem multi dimension record aggreg distribut data input lead fundament design data mine algorithm paper develop flexibl approach privaci preserv data mine requir problem specif algorithm map origin data set anonym data set anonym data close match characterist origin data includ correl dimens show extend method case data stream present empir result illustr effect method show effici method data stream correl burst event stream stock market data framework condens dbase anonym string data classifi data stream skew class distribut concept drift truth discoveri multipl conflict inform provid web world wide web import inform sourc guarante correct inform web web site provid conflict inform subject specif product paper propos problem call verac conform truth studi find true fact larg amount conflict inform subject provid web site design gener framework verac problem invent algorithm call truthfind util relationship web site inform web site trustworthi piec true inform piec inform true provid trustworthi web site experi show truthfind successfulli find true fact conflict inform identifi trustworthi web site popular search engin peertalk peer dto dpeer multiparti voic dover dip system top algorithm data mine abstract paper present top data mine algorithm identifi ieee intern confer data mine icdm decemb mean svm apriori em pagerank adaboost knn naiv bay cart top algorithm influenti data mine algorithm research commun algorithm provid descript algorithm discuss impact algorithm review current research algorithm algorithm cover classif leewav level dwise distribut wavelet coeffici process knn queri distribut stream fast monitor proxim central time devolv bipartit graph increment tensor analysi theori applic storag optim larg dscale distribut stream dprocess system dawn effici framework dct data error estim graph partit base link distribut exist graph partit approach base optim edg cut distribut edg weight link distribut consider paper propos gener model partit graph base link distribut model formul graph partit distribut assumpt approxim graph affin matrix distort measur model deriv graph partit algorithm approxim graph affin matrix bregman diverg correspond larg exponenti famili distribut establish connect edg cut object propos model provid unifi view graph partit ts dtree dalter search tree index trustworthi databas write donc dread dmani worm storag effici onlin tool search top dn gene similar biolog function gene ontolog databas temperatur dawar schedul singl dprocessor system abstract power awar oper system processor control ensur system temperatur exce threshold util system throttl clock speed scale equilibrium load denot constant polici compar zig zag polici altern phase cool heat paper character calcul zig zag polici argu simpl system throttl rule optim realiti system design forc implement zig zag polici case processor oper discret state requir altern cool heat phase set develop algorithm outperform zig zag polici present comput experi emphas perform algorithm object distinct distinguish object ident name sao stream index answer linear optim queri load shed framework optim dwai window stream join tupl drop commonli load shed stream oper inadequ window stream join join output rate overli reduc fail exploit time correl exist interrel stream paper introduc grubjoin adapt window stream join effect perform time correl awar cpu load shed grubjoin maxim output rate achiev nearoptim window harvest pick profit window segment join due combinatori explos join sequenc involv window segment window stream join pose uniqu challeng focu address quickli determin optim window harvest configur window stream join monitor learn time correl stream high accuraci minim overhead tackl challeng formal window harvest optim problem develop greedi heurist determin nearoptim window harvest configur approxim techniqu captur time correl experiment result show grubjoin vastli superior tupl drop time correl exist equal effect time correl nonexist adapt load diffus multiwai window stream join lira lightweight region dawar load shed mobil cq system provid high qualiti result locat base continu queri cq mobil system queri processor demand receiv frequent posit updat mobil node process frequent updat queri processor overload updat drop randomli bring queri result accuraci negat benefit frequent updat paper develop lira lightweight region awar load shed techniqu prevent reduc posit updat load queri processor maintain high qualiti queri result receiv updat randomli drop lira region awar partit mechan identifi benefici shed region cut posit updat mobil node region base densiti mobil node queri region lira judici appli amount updat reduct region aim minim neg impact load shed queri result accuraci experiment result show lira vastli superior random updat drop outperform altern possess region awar load shed capabl due lightweight natur lira introduc overhead load shed framework optim dwai window stream join tupl drop commonli load shed stream oper inadequ window stream join join output rate overli reduc fail exploit time correl exist interrel stream paper introduc grubjoin adapt window stream join effect perform time correl awar cpu load shed grubjoin maxim output rate achiev nearoptim window harvest pick profit window segment join due combinatori explos join sequenc involv window segment window stream join pose uniqu challeng focu address quickli determin optim window harvest configur window stream join monitor learn time correl stream high accuraci minim overhead tackl challeng formal window harvest optim problem develop greedi heurist determin nearoptim window harvest configur approxim techniqu captur time correl experiment result show grubjoin vastli superior tupl drop time correl exist equal effect time correl nonexist gstring approach effici search graph databas graph wide model complic data includ chemic compound protein interact xml document multimedia inform retriev data formul graph search problem find effici solut problem essenti applic popular approach repres graph queri graph sequenc convert graph search subsequ match state art sequenc method work finest granular node edg graph element result sequenc method semant consciou result sequenc bulki prone complex aris graph isomorph problem search paper introduc sequenc method captur semant underli graph data find meaning compon graph structur basic unit sequenc reduc size result sequenc enabl semant base search paper base approach chemic compound databas appli search complic graph protein structur experi demonstr approach outperform state ofth art graph search method mine coloss frequent pattern core pattern fusion extens research frequent pattern mine past decad brought number pattern mine algorithm effect effici exist frequent pattern mine algorithm encount challeng mine larg pattern call coloss frequent pattern presenc explos number frequent pattern coloss pattern critic applic domain bioinformat studi investig mine approach call pattern fusion effici find good approxim coloss pattern pattern fusion coloss pattern discov fuse small core pattern step increment pattern growth mine strategi adopt apriori fp growth examin larg number mid size properti distinguish pattern fusion exist frequent pattern mine approach draw mine methodolog empir studi show case current mine algorithm proce pattern fusion mine result set close approxim complet set coloss pattern qualiti evalu model propos paper commun learn graph approxim learn commun graph import problem domain type commun gener link pattern base commun paper propos gener model base graph approxim learn link pattern base commun structur graph model gener tradit graph partit approach applic learn commun structur model deriv famili algorithm flexibl learn commun structur easi incorpor prior knowledg commun structur experiment evalu theoret analysi show effect great potenti propos model algorithm effici discoveri frequent approxim sequenti pattern propos effici algorithm mine frequent approxim sequenti pattern ham distanc model algorithm gain effici adopt break build methodolog breakdown base observ occurr frequent pattern classifi group call strand develop effici algorithm quickli mine strand iter growth build stage strand group form support set approxim pattern identifi salient featur algorithm abil grow frequent pattern iter assembl build block signific size local search fashion avoid increment growth global search achiev greater effici lose complet mine result experiment studi demonstr algorithm effici mine global repeat approxim sequenti pattern miss exist method dmanag media stream process servic overlai relat cluster symmetr convex code relat data frequent machin learn applic relat data consist pairwis relat similar dissimilar pair implicit object store relat matric typic knowledg relat cluster formul graph partit applic formul adequ gener relat data paper propos gener model relat cluster base symmetr convex code model applic type relat data unifi exist graph partit formul model deriv altern bound optim algorithm solv symmetr convex code popular distanc function euclidean distanc gener diverg experiment evalu theoret analysi show effect great potenti propos model algorithm bridgenet adapt multi dsourc stream dissemin overlai network graph dbase approach systemat reconstruct human transcript regulatori modul storag optim larg dscale distribut stream process system graph dbase approach systemat reconstruct human transcript regulatori modul truth discoveri multipl conflict inform provid web world wide web import inform sourc guarante correct inform web web site provid conflict inform subject specif product paper propos problem call verac conform truth studi find true fact larg amount conflict inform subject provid web site design gener framework verac problem invent algorithm call truthfind util relationship web site inform web site trustworthi piec true inform piec inform true provid trustworthi web site experi show truthfind successfulli find true fact conflict inform identifi trustworthi web site popular search engin time ddepend event hierarchi construct string classif data stream string data recent import number applic comput molecular biologi protein analysi market basket data case string wide varieti substructur physic signific applic substructur repres import fragment dna string interest portion fraudul transact case desir determin ident locat extent substructur data difficult gener classif problem problem label entir string deal complex task determin string fragment kind behavior problem complic kind substr show complic nest pattern defin problem refer gener classif problem propos scalabl approach base hidden markov model problem show implement gener string classif procedur larg data base data stream present experiment result number larg data set data stream probabilist framework relat cluster relat cluster attract attent due phenomen impact import applic involv multi type interrel data object web mine search market bioinformat citat analysi epidemiolog paper propos probabilist model relat cluster princip framework unifi import cluster task includ tradit attribut base cluster semi supervis cluster cluster graph cluster propos model seek identifi cluster structur type data object interact pattern type object model propos parametr hard soft relat cluster algorithm larg number exponenti famili distribut algorithm applic relat data structur time unifi number stat art cluster algorithm cluster algorithm partit graph cluster semi supervis cluster base hidden markov random field graphscop paramet dfree mine larg time devolv graph gprune constraint push framework graph pattern mine abstract graph mine applic increasingli strong urg impos user constraint mine result unlik tradit itemset constraint structur constraint densiti diamet graph hard push deep mine process paper give comprehens studi prune properti tradit structur constraint aim reduc pattern search space data search space gener framework call gprune propos incorpor constraint recurs reinforc entir mine process concept pattern insepar data antimonoton propos handl structur constraint uniqu context graph combin prune properti comprehens unifi classif framework structur constraint explor antimonoton context graph pattern mine signific extens classif constraint deepen understand prune properti structur graph constraint visual explor genom data widespread accept uptak grid technolog achiev ensur secur mechan need support grid base collabor strong local secur mechan predomin secur address grid commun public kei infrastructur pki support authent whilst pki address user ident issu authent provid fine grain control user allow remot resourc authoris grid commun put forward numer softwar propos authoris infrastructur akenti ca cardea gsi permi vom clear forese futur collect solut norm address global grid forum ggf propos gener saml base authoris api principl fine grain control authoris access grid servic experi appli stress test api varieti applic domain essenti give insight practic aspect larg scale usag authoris infrastructur paper present experi dti fund bridg project jisc fund dyvos project api globu version permi authoris infrastructur gener framework mine concept ddrift data stream skew distribut recent year interest studi predict model data stream studi assum balanc stabl data stream handl skew posit lot neg stochast distribut typic data stream applic paper propos approach mine data stream estim reliabl posterior probabl ensembl model match distribut sampl neg repeat sampl posit formal show interest import properti propos framework reliabl estim probabl skew posit class accuraci estim probabl effici scalabl experi perform synthet real world dataset skew distribut demonstr framework substanti advantag exist approach estim reliabl predic accuraci system keyword search textual stream system keyword search textual stream anonym string data blink rank keyword search graph queri process graph structur data enjoi grow number applic top keyword search queri graph nd top answer rank criteria answer substructur graph queri keyword current techniqu support queri gener graph suffer drawback poor worst case perform take full advantag index high memori requir address problem propos blink bi level index queri process scheme top keyword search graph blink search strategi provabl perform bound addition exploit bi level index prune acceler search reduc index space blink partit data graph block bilevel index store summari inform block level initi guid search block detail inform block acceler search block experi show blink offer order magnitud perform improv exist approach resourc dadapt real dtime event detect challeng experi prototyp multi dmodal stream analyt monitor applic system graph contain search index set model graph queri graph contain search aim find model graph due wide adopt graph model fast contain search graph data find applic domain comparison tradit graph search retriev graph contain search index characterist examin paper perform systemat studi characterist propos contrast subgraph base index model call cindex contrast subgraph captur structur differ model graph queri graph perfect index due high select redund awar featur select process cindex sort set signific distinct contrast subgraph maxim index capabl show np complet choos set index featur greedi algorithm approxim level optim index ratio take solut base index model extend accommod hierarch index methodolog appli data space cluster sampl techniqu reduc index construct time propos methodolog gener solut contain search index graph data transit relat experiment result real test data show cindex achiev optim prune power contain search workload confirm obviou advantag indic built tradit graph search scenario cellsort high perform sort cell processor paper describ design implement cellsort high perform distribut sort algorithm cell processor design cellsort distribut biton merg data parallel biton sort kernel order exploit architectur cell processor make form parallel achiev good scalabl structur cellsort tier sort tier simd singl instruct multipl data optim biton sort sort kb item cat fit spe processor cell local store design comprehens simdiz scheme emploi data parallel fine grain step biton sort kernel result show simdiz biton sort kernel vastli superior altern spe perform time faster compar quick sort ghz intel xeon tier core biton merg optim cross spe data transfer asynchron dma sort number item fit cumul space local store particip spe design data transfer synchron patter minim serial section code take advantag high aggreg cross spe bandwidth cell result show core biton sort scale cell processor increas number spe perform time faster spe compar parallel quick sort dual ghz intel xeon tier core biton merg sort larg number item store main memori result show properli implement distribut core biton sort cell significantli outperform asymptot averag case superior quick sort larg number memori resid item time faster sort gb data spe compar dual ghz intel xeon term core impli disk base sort context paper relat extern sort strong section detail execut stream join cell processor low latenc high throughput process kei requir data stream manag system dsmss multi core processor provid high aggreg process capac ideal match execut costli dsm oper recent develop cell processor good heterogen multi core architectur power platform execut data stream oper high perform side exploit full potenti multi core processor cell challeng due heterogen natur process element softwar manag local memori processor side unconvent program model gener paper studi problem scalabl execut window stream join oper multi core processor specif cell processor examin aspect join execut flow determin set techniqu appli order minim sequenti segment maxim parallel concret show basic window coupl low overhead pointer shift techniqu achiev effici join window partit column orient join window organ minim scatter data transfer delai optim doubl buffer effect pipelin rateawar batch balanc join throughput tupl delai final simd singl instruct multipledata optim oper code exploit data parallel experiment result show design guidelin implement techniqu outlin paper window stream join achiev high scalabl linear number processor make effici extens hardwar parallel provid cell processor reach data process rate gb sec significantli surpass perform obtain form convent high end processor support combin input stream rate tupl sec minut window drop tupl result time higher output rate compar sse implement dual ghz intel xeon time seri compress privaci paper studi trade off time seri compress partial inform hide fundament implic introduc uncertainti individu valu perturb specif perturb compress properti origin data detect filter reduc uncertainti make perturb similar origin data preserv structur data simultan make breach harder data compress fraction uncertainti remov true valu leak reveal perturb formal notion studi trade off real data develop practic scheme strike good balanc extend fly data hide stream environ graph contain search index set model graph queri graph contain search aim find model graph due wide adopt graph model fast contain search graph data find applic domain comparison tradit graph search retriev graph contain search index characterist examin paper perform systemat studi characterist propos contrast subgraph base index model call cindex contrast subgraph captur structur differ model graph queri graph perfect index due high select redund awar featur select process cindex sort set signific distinct contrast subgraph maxim index capabl show np complet choos set index featur greedi algorithm approxim level optim index ratio take solut base index model extend accommod hierarch index methodolog appli data space cluster sampl techniqu reduc index construct time propos methodolog gener solut contain search index graph data transit relat experiment result real test data show cindex achiev optim prune power contain search workload confirm obviou advantag indic built tradit graph search scenario graph index tree delta grap distribut infrastructur support person servic mobil web method measur semant similar term crossclu user dguid multi drelat cluster domain ddriven action knowledg discoveri grubjoin adapt tupl drop commonli load shed stream oper inadequ window stream join join output rate overli reduc fail exploit time correl exist interrel stream paper introduc grubjoin adapt window stream join effect perform time correl awar cpu load shed grubjoin maxim output rate achiev nearoptim window harvest pick profit window segment join due combinatori explos join sequenc involv window segment window stream join pose uniqu challeng focu address quickli determin optim window harvest configur window stream join monitor learn time correl stream high accuraci minim overhead tackl challeng formal window harvest optim problem develop greedi heurist determin nearoptim window harvest configur approxim techniqu captur time correl experiment result show grubjoin vastli superior tupl drop time correl exist equal effect time correl nonexist approxim queri process cube stream anonym classif data privaci preserv abstract classif fundament problem data analysi train classifi requir access larg collect data releas person specif data custom data patient record pose threat individu privaci remov explicit identifi inform ssn link releas record back ident match combin identifi attribut sex zip birthdat approach combat link attack call anonym anonym link attribut releas record match combin link attribut previou work attempt find optim anonym minim data distort metric argu minim distort train data relev classif goal requir extract structur predic futur data paper propos anonym solut classif goal find anonym necessarili optim sens minim data distort preserv classif structur conduct intens experi evalu impact anonym classif futur data experi real life data show qualiti classif preserv highli restrict anonym requir index term privaci protect anonym secur integr data mine classif data share asap adapt sampl approach data collect sensor network abstract promin comprehens wai data collect sensor network period extract raw sensor read data collect enabl complex analysi data network aggreg queri process flexibl data analysi cost power consumpt paper develop asap adapt sampl approach energyeffici period data collect sensor network main idea asap dynam chang subset node sampler sensor read sampler node directli collect valu sampler node predict probabilist model local period construct asap effect increas network lifetim keep qualiti collect data high scenario fragment proxi cach stream multimedia object handicap attack confid altern danonym cpu load shed binari stream join dddm domain driven data mine proceed acm cikm intern confer inform knowledg manag advanc web mine web usag analysi classif span privat databas vico adapt distribut video correl system emerg applic video sensor monitor benefit line video correl system discov linkag video stream realtim line video correl resourc intens singl host easili overload present adapt distribut line video correl system call vico unlik singl stream process correl video stream requir distribut execut system observ correl constraint correl data distribut host vico achiev uniqu featur correl awar vico guarante correl accuraci spread excess workload multipl host adapt system adjust algorithm behavior switch algorithm adapt dynam stream environ finegranular workload resourc intens correl request divid distribut multipl host implement deploi prototyp vico commerci cluster system experi result real video synthet workload show vico outperform exist techniqu scale perform video correl continu keyword search multipl text stream paper address issu continu keyword queri multipl textual stream explor techniqu extract inform paper repres knowledg approach perform keyword search multipl textual stream scenario intuit assum research financi analyst search inform topic continu poll data multipl possibl heterogen text stream rss feed blog topic interest resourc adapt period estim stream data multi dqueri sql progress indic abstract recent progress indic propos sql queri rdbmss previous propos progress indic queri isol ignor impact simultan run queri perform paper explor multi queri progress indic explicitli consid concurr run queri queri predict arriv futur produc estim demonstr multi queri progress indic provid accur estim singl queri progress indic extend progress indic gui tool show appli multi queri progress indic workload manag report initi implement multi queri progress indic postgresql experi estim remain queri execut time workload manag fast comput reachabl label larg graph numer applic deal larg graph queri reachabl node graph hop cover compactli repres edg transit closur graph space answer reachabl queri effici challeng comput hop cover exist approach suffer larg resourc consumpt low compress rate paper propos hierarch partit approach partit larg graph subgraph repeatedli top fashion uniqu featur approach comput hop cover partit iter top partit provid techniqu comput hop cover connect subgraph cover comput cut graph subgraph result cover high compress entir graph approach propos node orient approach edg orient approach approach effici comput hop cover larg graph high compress rate extens experi studi show hop cover graph node billion connect obtain minut compress rate pc mine index similar search graph complex structur dual label answer graph reachabl queri constant time graph reachabl fundament wide rang applic includ xml index geograph navig internet rout ontolog queri base rdf owl applic involv huge graph requir fast answer reachabl queri reachabl label method propos purpos assign label vertic reachabl vertic decid label spars graph hop base reachabl label scheme answer reachabl queri effici small label space label process time consum practic larg graph paper propos label scheme spars graph scheme ensur graph reachabl queri answer constant time spars graph complex label process linear make algorithm applic massiv dataset analyt experiment result show approach effici stateof art approach label method altern scheme tradeoff queri time label space benefit applic tree graph search substructur superimpos distanc effici index techniqu develop exact approxim substructur search larg scale graph databas retriev problem structur categor geometr distanc constraint solv paper develop method call pi partit base graph index search support similar search substructur superimpos distanc constraint pi select discrimin fragment queri graph index prune graph violat distanc constraint identifi criterion distinguish select fragment multipl graph develop partit method obtain set highli select fragment improv prune perform experiment result show pi effect process real graph queri window dbase tensor analysi high ddimension multi daspect stream data stream valu multipl aspect environment sensor type temperatur humid locat timestamp type locat addit aspect model stream simultan find pattern multipl aspect increment stream fashion paper problem address gener data model tensor stream effect algorithm framework window base tensor analysi wta variat wta independentwindow tensor analysi iw move window tensor analysi mw present evalu extens real dataset final illustr import applic multi aspect correl analysi maca wta demonstr effect environment monitor applic local correl track time seri address problem captur track local correl time evolv time seri approach base compar local auto covari matric spectral decomposit seri gener notion linear cross correl concis captur wide varieti local pattern trend method produc gener similar score evolv time accur reflect chang relationship final estim increment stream set demonstr us robust effici wide rang real dataset discov partial order binari data find talk voip network progress stream cluster loci load shed class dpreserv data acquisit ac dclose effici mine approxim close itemset core pattern recoveri recent studi propos method discov approxim frequent itemset presenc random nois relax rigid requir exact frequent pattern mine interest pattern previous fragment exact pattern mine method due random nois measur error successfulli recov larg number uninterest candid explor mine process result relax pattern mine methodolog sever slow mine process wors hard end user distinguish recov interest pattern uninterest paper propos effici algorithm ac close recov approxim close itemset core pattern focus call core pattern integr top mine effect prune strategi algorithm narrow search space potenti interest experiment result show ac close substanti outperform previous propos method term effici deliv similar set interest recov pattern spectral cluster multi dtype relat data cluster multi type relat data attract attent recent year due high impact import applic web mine commerc bioinformat research gener multi type relat data cluster limit preliminari contribut paper fold propos gener model collect factor relat matric multi type relat data cluster model applic relat data structur model deriv algorithm spectral relat cluster cluster multi type interrel data object simultan algorithm iter emb type data object low dimension space benefit interact hidden structur type data object extens experi demonstr promis effect propos algorithm show exist spectral cluster algorithm consid special case propos model algorithm demonstr good theoret gener propos model algorithm gener framework accur fast regress data summar random decis tree global distanc dbase segment trajectori work introduc distanc base criteria segment object trajectori segment lead simplif origin object smaller complex primit suit storag retriev purpos previou work trajectori segment attack problem local segment separ trajectori databas directli optim inter object separ mine oper search cluster classif larg databas paper analyz trajectori segment problem global perspect util data awar distanc base optim techniqu optim pairwis distanc estim lead effici object prune deriv exact solut distanc base formul due intract complex exact solut present approxim greedi solut exploit forward search local optim solut greedi solut impos prohibit comput cost put forward lightweight varianc base segment techniqu intellig relax pairwis distanc area affect mine oper suppress model overfit mine concept ddrift data stream gplag detect softwar plagiar program depend graph analysi blossom open sourc project conveni softwar plagiar compani disciplin tempt plagiar open sourc project product current plagiar detect tool suffici academ short fight plagiarist disguis statement reorder code insert effect confus tool paper develop plagiar detect tool call gplag detect plagiar mine program depend graph pdg pdg graphic represent data control depend procedur pdg invari plagiar gplag effect state art tool plagiar detect order make gplag scalabl larg program statist lossi filter propos prune plagiar search space experi studi show gplag effect effici detect plagiar easili slip exist tool take second find simul plagiar program thousand line code riva index visual high ddimension data dimens reorder discov overlap commun name entiti abstract commun discoveri base social network analysi studi extens web hyperlink environ limit research case name entiti text document cooccurr entiti document impli connect investig connect reveal import pattern paper mine commun name entiti web document text corpu exist work commun discoveri gener partit entiti network assum entiti belong commun scenario name entiti entiti particip commun person commun famili colleagu friend paper propos techniqu mine overlap commun name entiti techniqu base triangl format expans cluster content similar experiment result show propos techniqu highli effect framework resourc dawar knowledg discoveri data stream holist approach applic cluster framework cluster massiv text categor data stream applic new group filter text crawl document organ requir real time cluster segment text data record categor data stream cluster problem number applic problem custom segment real time trend analysi present onlin approach cluster massiv text categor data stream statist summar methodolog present result illustr effect techniqu optim multi dscale pattern time seri stream introduc method discov optim local pattern concis describ main trend time seri approach examin time seri multipl time scale window size effici discov kei pattern introduc criterion select window size concis captur kei oscillatori aperiod trend kei insight li learn optim orthonorm transform data oppos predetermin basi approxim function piecewis constant shortwindow fourier wavelet essenti restrict famili trend method lift limit lend fast increment estim stream set experiment evalu show method captur meaning pattern varieti set stream approach requir order magnitud time space produc concis inform pattern linkclu effici cluster heterogen semant link data object relat databas cross link multi type link link rich semant inform import relationship object current cluster method reli properti belong object se similar object link desir cluster gener properti object paper explor linkag base cluster similar object measur base similar object link comparison previou studi simrank comput link recurs pair object advantag power law distribut link develop hierarch structur call simtre repres similar multi granular manner method avoid high cost comput store pairwis similar explor relationship object effici algorithm propos comput similar object avoid pairwis similar comput merg comput branch simtre experi show propos approach achiev high effici scalabl accuraci cluster multi type link object ddemand index effici structur join featur dbase similar search graph structur similar search complex structur import oper graph relat applic exact match restrict articl investig issu substructur similar search index featur graph databas transform edg relax ratio queri graph maximum allow featur miss structur filter algorithm filter graph perform pairwis similar comput shown featur result poor filter perform challeng design effect featur set select strategi maxim filter capabl prove complex optim featur set select worst case number featur select practic identifi criteria build effect featur set filter demonstr combin featur similar size select improv filter search perform significantli multi filter composit framework propos featur base filter concept gener appli search approxim consecut sequenc tree structur data categori subject descriptor databas manag system queri process preliminari releas articl accept acm transact databas system definit version product acm releas supersed version structur period measur time dseri data text classif neg exampl revisit discov frequent close partial order string abstract mine knowledg order sequenc data import problem applic bioinformat web mine network manag intrus detect custom follow partial order purchas seri product partial order predict relat custom futur purchas develop market campaign biolog sequenc microarrai data cluster base partial order share sequenc set item total order subset item repres string string databas multiset string paper identifi problem mine frequent close partial order string frequent close partial order captur nonredund interest order inform string databas importantli mine frequent close partial order discov meaning knowledg disclos previou data mine techniqu problem mine frequent close partial order challeng tackl problem develop frecpo frequent close partial order practic effici algorithm mine complet set frequent close partial order larg string databas interest prune techniqu devis speed search report extens perform studi real data set synthet data set illustr effect effici approach index term frequent pattern close pattern partial order string data mine framework ddemand classif evolv data stream increment process continu rang queri move object framework ddemand classif evolv data stream process move queri move object motion dadapt index abstract paper describ motion adapt index scheme effici evalu move continu queri mcq move object concept motion sensit bound box msb model move object move queri bound box automat adapt size dynam motion behavior individu object index frequent chang object posit index frequent chang object queri msb updat bound box need object queri move boundari box help decreas number updat index importantli predict queri result optimist precalcul queri result decreas number search index motion sensit bound box increment updat predict queri result introduc concept guarante safe radiu optimist safe radiu extend motion adapt index scheme evalu move continu nearest neighbor knn queri experi show propos motion adapt index scheme effici evalu move continu rang queri move continu knn queri index term move object databas spatio tempor index continu queri effici classif multipl databas relat crossmin approach detect classif evolv data stream catch moment maintain close frequent itemset data stream slide window paper consid problem mine close frequent itemset data stream slide window limit memori space design synopsi data structur monitor transact slide window output current close frequent itemset time due time memori constraint synopsi data structur monitor itemset monitor frequent itemset make imposs detect itemset frequent paper introduc compact data structur close enumer tree cet maintain dynam select set itemset slide window select itemset boundari close frequent itemset rest itemset concept drift data stream reflect boundari movement cet word statu chang itemset frequent frequent occur boundari queri index contain dencod interv effici stream process holist approach resourc dawar adapt data stream mine support multi dparti voic dover dip servic peer dto dpeer stream process multi parti voic ip mvoip servic provid econom natur group commun mechan emerg applic line game distanc collabor tele immers paper present peer peer stream process system call peertalk provid resourc effici failur resili mvoip servic previou work solut fulli distribut organ requir special server ip multicast support decoupl stream process mvoip servic phase aggreg phase mix audio stream activ speaker singl stream distribut phase distribut mix audio stream listen decoupl model optim adapt stream mix distribut process separ specif adapt spread stream mix workload resourc constrain peer host current speak activ implement prototyp peertalk system conduct experi real world wide area network result show peertalk achiev lower resourc content servic qualiti previou common solut rotat invari index shape line draw adapt load shed window stream join present adapt load shed approach window stream join contrast convent approach drop tupl input stream explor concept select process load shed stream tupl store window shed excess cpu load perform join oper entir set tupl window dynam chang subset tupl learn highli benefici support dynam select process form runtim adapt adapt input stream rate adapt time correl stream adapt join direct load shed approach enabl integr util base load shed time correl base load shed index speed execut stream join experi conduct evalu adapt load shed term output rate util result show select process approach load shed effect significantli outperform approach drop tupl input stream integr dct dwt approxim cube stream compact reachabl label graph dstructur data test reachabl node graph problem import applic includ knowledg represent program analysi recent biolog ontolog databas inferenc xml queri process approach propos encod graph reachabl inform node label scheme exist scheme work specif type graph paper propos approach hlss hybrid label structur identifi type substructur graph encod techniqu suitabl characterist implement hlss effici phase algorithm phase identifi encod strongli connect compon tree substructur phase encod remain reachabl relationship compress dens rectangular submatric transit closur matrix import subproblem find densest submatric demonstr hard problem propos practic algorithm experi show hlss handl type graph exist approach fall prei graph substructur design handl final discuss updat reachabl label graph updat qualit show hlss support effici updat exist approach data stream mine resourc adapt comput optim compon composit scalabl stream process stream process increasingli import emerg stream applic audio video surveil stock price trace sensor data analysi challeng problem provid optim compon composit distribut stream process environ goal optim compon composit achiev load balanc subject multipl function resourc qualiti servic qo constraint compos stream applic paper present adapt composit probe acp approach problem previou work acp hybrid approach combin distribut composit probe coars grain global state manag guid coars grain global state inform acp select probe subset candid compon discov approxim optim compon composit acp tune adapt adjust number probe maintain composit perform target composit success rate dynam stream environ optim compon composit problem np hard acp approach adapt polynomi approxim solut conduct extens simul experi show effici scalabl adapt acp approach compar altern solut onlin mine data stream applic techniqu progress top ddown special inform privaci preserv releas person specif data specif state pose threat individu privaci paper present practic effici algorithm determin gener version data mask sensit inform remain model classif gener data implement special detail level inform top manner minimum privaci requir violat top special natur effici handl categor continu attribut approach exploit fact data redund structur classif gener elimin structur structur emerg result show qualiti classif preserv highli restrict privaci requir work great applic public privat sector share inform mutual benefit product text classif label neg document effect estim posterior probabl explain accuraci random decis tree approach combin multipl cluster soft correspond border dbase approach hide sensit frequent itemset templat dbase privaci preserv classif problem privaci preserv import issu releas data mine purpos anonym model introduc protect individu identif recent studi show sophist model protect associ individu sensit inform paper propos anonym model protect identif relationship sensit inform data discuss properti anonym model prove optim anonym problem np hard present optim globalrecod method anonym problem propos local recod algorithm scalabl result data distort effect effici shown experi describ model extend gener case improv categor classifi sensit sampl select bia recent paper categor classifi learn algorithm sensit common type sampl select bia chanc select train sampl depend featur vector directli class label classifi learner categor local insensit type sampl select bia consid global paper true model distinguish model algorithm output discuss bayesian classifi logist regress hard margin support vector machin true model model gener true class label implicitli assum contain model space learner true class probabl model estim class probabl assum asymptot converg train data set size increas discuss naiv bay decis tree soft margin support vector machin model space assum true model classif algorithm argu global learner argu classifi learn algorithm includ discuss affect sampl select bia depend dataset heurist induct bia impli classifi learn algorithm appropri dataset make earlier experiment result produc addit result illustr claim focus commun discoveri commun discoveri base social network studi extens web hyperlink environ limit research case web document occurr word entiti sentenc document impli connect studi connect reveal import relationship paper investig cooccurr name entiti web page blog mine commun entiti show identifi commun environ transform graph cluster problem hierarch cluster algorithm propos exploit triangl structur graph mutual inform vertic empir studi show propos algorithm promis discov commun web document effici mine frequent close partial order csr speaker recognit compress voip packet stream mine evolv stream resourc adapt comput cross drelat cluster user guidanc dcluster block decomposit collabor filter aim predict test user rate item integr mind user rate inform tradit collabor filter method suffer fundament problem sparsiti scalabl paper propos framework collabor filter appli orthogon nonneg matrix tri factor onmtf allevi sparsiti problem matrix factor solv scalabl problem simultan cluster row column user item matrix experiment result benchmark data set present show algorithm toler sparsiti scalabl achiev good perform pattern dbase similar search microarrai data adapt load diffus stream join abstract data stream process increasingli import emerg applic call sophist realtim process data stream stock trade surveil network traffic monitor sensor data analysi stream join import stream process oper detect linkag correl data stream major challeng process stream join handl continu high volum time vari data stream resourc constraint paper present load diffus system enabl scalabl execut resourc intens stream join ensembl server host load diffus achiev simpl correl awar stream partit algorithm previou work load diffus system achiev fine grain load share distribut stream process system produc exact queri answer miss join result gener duplic join result experiment result show load diffus scheme greatli improv system throughput achiev balanc load distribut increment process continu rang queri locat dawar servic applic multi dmetric index euclidean period match fast burst correl financi data abstract examin problem monitor identif correl burst pattern multi stream time seri databas methodolog compris step burst detect part burst index step burst detect scheme impos variabl threshold examin data take advantag skew distribut typic encount applic index step util memori base interv index effect identifi overlap burst region focu work financi data propos method data structur find applic anomali novelti detect telecommun network traffic medic data final manifest real time respons burst index techniqu demonstr us approach correl surpris volum trade event ny stock exchang dlevel distribut architectur effici web content adapt deliveri build decis tree record link kei refer classif problem data collect tabl relat hierarch structur kei refer class label contain root tabl parent tabl repres relationship type child tabl data frequent found relat databas data warehous xml data biolog databas solut join tabl univers tabl base record relationship suffer signific blowup caus relationship solut treat problem relat learn cost increas complex degrad perform propos method build decis tree classifi built join tabl blowup requir tradit approach iter mine rule constrain anteced loadstar load shed scheme classifi data stream problem resourc alloc mine multipl data stream due larg volum high speed stream data mine algorithm cope ect system overload realiz maximum mine benefit resourc constraint challeng task paper propos load shed scheme classifi multipl data stream focu problem classifi data drop load shed scheme ii decid drop data stream introduc qualiti decis qod metric measur level uncertainti classif exact featur valu data load shed markov model predict distribut featur valu make classif decis predict valu qod metric resourc alloc multipl data stream maxim qualiti classif decis load shed scheme learn adapt chang data characterist data stream experi synthet data real life data show load shed scheme ectiv improv accuraci classif resourc constraint keyword data mine data stream load shed classif qualiti decis featur predict markov model mine behavior graph backtrac noncrash bug analyz execut buggi softwar program essenti data mine process interest method develop trace crash bug memori violat core dump difficult analyz noncrash bug logic error paper develop method classifi structur trace program execut softwar behavior graph analyz correct incorrect execut made good progress isol program region lead faulti execut classif framework built integr close graph mine svm classif interestingli suspici region identifi captur classif accuraci chang measur increment program execut perform studi case base experi show approach effect effici dneighbor search pattern distanc space period detect structur period similar variabl constraint privaci preserv data mine onlin analysi commun evolut data stream push featur select ahead join abstract current approach featur select multipl data sourc join data order evalu featur class label scalabl involv unnecessari inform leakag paper present perform featur select class propag elimin join featur select propag compact data structur inform select featur data sourc allow featur evalu local inform experi confirm algorithm highli scalabl effect preserv data privaci substructur similar search graph databas advanc databas system face great challeng rais emerg massiv complex structur data bioinformat chem informat applic fundament support need applic effici search complex structur data exact match restrict similar search complex structur vital oper support effici paper investig issu substructur similar search index featur graph databas transform edg relax ratio queri graph maximum allow miss featur structur filter algorithm call grafil filter graph perform pairwis similar comput shown featur result poor filter perform challeng design effect featur set select strategi filter examin effect featur select mechan develop multi filter composit strategi filter distinct complementari subset featur identifi criteria form effect featur set filter demonstr combin featur similar size select improv filter search perform significantli concept present grafil appli search approxim consecut sequenc tree complic structur loadstar load shed data stream mine demo show intellig load shed essenti achiev optimum result mine data stream resourc constraint loadstar system introduc load shed techniqu classifi multipl data stream larg volum high speed loadstar metric qualiti decis qod measur level uncertainti classif resourc alloc sourc uncertainti high make optimum classif decis accur qod measur loadstar reli featur predict model data drop load shed mechan loadstar adapt chang data characterist data stream system offer nice solut data mine resourc constraint paramet free bursti event detect text stream cluster techniqu chang diagnosi data stream effici structur join dthe dfly index previou work structur join focus maintain in index disk requir element set sort paper studi fly memori index approach structur join sort element maintain index disk identifi similar structur join problem stab queri problem extend main memori base index techniqu stab queri structur join ad tempor dimens search case studi public search mine data stream abstract mine data stream rais number research challeng data mine commun challeng includ limit comput resourc mine stream data mobil devic limit resourc due continu data stream algorithm pass incom data record articl algorithm output granular aog approach mine data stream discuss aog adapt approach cope challeng inher featur data stream show result aog base cluster resourc constrain environ high dimension project cluster data stream effici process continu rang queri locat dawar mobil servic improv biclust method analyz gene express profil webkdd web mine web usag analysi post dworkshop report demand ddriven frequent itemset mine pattern structur webkdd web mine web usag analysi post dworkshop report effect effici algorithm high ddimension outlier detect proceed acm sigmod workshop research issu data mine knowledg discoveri text classif label word tradition text classifi built label train exampl label manual human expert user labor intens time consum process past year research investig form semi supervis learn reduc burden manual label paper propos approach label set document propos method label set repres word class word extract set document class set unlabel document form initi train set em algorithm appli build classifi kei issu approach obtain set repres word class user provid difficult user give word insuffici accur learn propos method solv problem combin cluster featur select techniqu effect rank word unlabel set import user select label word rank list class process requir effort provid word manual label document result show method highli effect promis monitor continu rang queri paper propos gener framework monitor continu spatial queri move object framework distinguish exist work address locat updat issu provid common interfac monitor mix type queri base notion safe region client locat updat strategi develop base queri monitor significantli reduc wireless commun queri reevalu cost requir maintain upto date queri result propos algorithm queri evalu reevalu safe region comput framework enhanc propos advantag practic mobil assumpt maximum speed steadi movement experiment result show framework substanti outperform tradit period monitor scheme term monitor accuraci cpu time achiev close optim wireless commun cost framework scale larg monitor system robust object mobil pattern shingl dbase queri index locat dbase mobil dcommerc motion adapt index move continu queri move object paper describ motion adapt index scheme effici evalu move continu queri mcq move object concept motion sensit bound box msb model move object move queri bound box automat adapt size dynam motion behavior individu object index frequent chang object posit index frequent chang object queri msb updat bound box need object queri move boundari box help decreas number updat index importantli predict queri result optimist precalcul queri result decreas number search index motion sensit bound box increment updat predict queri result experi show propos motion adapt index scheme effici evalu move continu rang queri interv queri index effici stream process crossmin effici classif multipl databas relat todai structur data store relat databas databas consist multipl relat link conceptu entiti relationship link design relat databas schema multi relat classif wide disciplin financi decis make medic research geograph applic classif approach work singl flat data relat difficult convert multipl relat singl flat relat introduc huge undesir univers relat lose essenti inform previou work induct logic program approach recent relat mine proven effect high accuraci multi relat classif suffer poor scalabl number relat number attribut databas condens approach privaci preserv data mine issu research report earli dissemin content view transfer copyright publish distribut ibm prior public limit peer commun specif request public request fill reprint legal obtain copi articl payment royalti copi request ibm watson research center mine extrem skew trade anomali index continu rang queri locat dawar mobil servic index continu rang queri cover tile fast locat move object crossmin effici classif multipl databas relat todai structur data store relat databas databas consist multipl relat link conceptu entiti relationship link design relat databas schema multi relat classif wide disciplin financi decis make medic research geograph applic classif approach work singl flat data relat difficult convert multipl relat singl flat relat introduc huge undesir univers relat lose essenti inform previou work induct logic program approach recent relat mine proven effect high accuraci multi relat classif suffer poor scalabl number relat number attribut databas bottom dup gener data mine solut privaci protect privaci preserv data mine modifi exist data mine techniqu random data paper investig data mine techniqu mask data term data mine base privaci protect approach incorpor partial requir target data mine task process mask data essenti structur preserv mask data idea simpl explor data gener concept data mine hide detail inform discov trend pattern data mask standard data mine techniqu appli modif work demonstr posit data mine technolog discov pattern mask privat inform privaci problem data holder releas version data build classif model protect link releas data extern sourc infer sensit inform adapt iter bottom gener data mine gener data gener data remain classif difficult link sourc gener space hierarch structur gener kei identifi gener climb hierarchi iter enumer candid gener impract present scalabl solut examin gener iter attribut involv link decis tree evolut limit number label data item drift data stream previous propos mine method data stream make unrealist assumpt label data stream readili mine anytim real world problem label data stream rare immedi due reason model reconstruct label data period passiv stream mine model drawback propos concept demand driven activ data mine activ mine loss model continu guess true class label estim small number instanc actual class label verifi pai afford cost estim loss toler threshold model evolv small number instanc verifi true class label previou work activ mine concentr error guess estim paper discuss approach decis tree evolut moment maintain close frequent itemset stream slide window paper consid problem mine close frequent itemset slide window limit memori space design synopsi data structur monitor transact slide window output current close frequent itemset time due time memori constraint synopsi data structur monitor itemset monitor frequent itemset make imposs detect itemset frequent paper introduc compact data structur close enumer tree cet maintain dynam select set itemset slide window select itemset consist boundari close frequent itemset rest itemset concept drift data stream reflect boundari movement cet word statu chang itemset frequent frequent occur boundari boundari stabl cost mine close frequent itemset slide window dramat reduc mine transact possibl boundari movement cet experi show algorithm perform previou approach demand classif data stream current model classif problem ectiv handl burst class come time fact current model classif problem simpli concentr method pass classif model larg data set model data stream classif view data stream classif problem point view dynam approach simultan train test stream dynam classif data set model reflect real life situat ectiv desir classifi test stream real time evolv train test stream aim creat classif system train model adapt quickli underli data stream order achiev goal propos demand classif process dynam select window past train data build classifi empir result system maintain high classif accuraci evolv data stream provid cient solut classif task analysi peer dto dpeer system workload character effect traffic cacheabl process continu rang queri move object vcr dbase queri index mine evolv data stream privaci preserv privaci preserv data mine research direct data mine statist databas data mine algorithm analyz side effect incur data privaci studi effici discoveri frequent itemset privaci preserv data mine nontrivi maintain discov frequent itemset databas frequent itemset updat frequent itemset turn infrequ itemset paper increment updat algorithm ippfim propos effici mainten discov frequent itemset transact data ad transact databas privaci preserv algorithm make previou mine result cut cost find frequent itemset updat databas perform evalu show effici method keyword data mine privaci preserv increment vcr index fast event match highli doverlap rang predic activ mine data stream previous propos mine method data stream make unrealist assumpt label data stream readili mine anytim real world problem label data stream rare immedi due reason model reconstruct label data period passiv stream mine model drawback propos concept demand driven activ data mine activ mine loss model continu guess true class label estim small number instanc actual class label verifi pai afford cost estim loss toler threshold model evolv small number instanc verifi true class label previou work activ mine concentr error guess estim paper discuss approach decis tree evolut resourc dawar mine variabl granular data stream data stream applic approxim adapt import issu effect mine explor paper fundament problem limit resourc memori space comput power util produc accur estim import featur track mine pattern properli util resourc examin issu tempor granular refer phenomenon time advanc peopl interest recent event mean resourc util explor recent data finer granular mine task discov frequent tempor pattern resourc expect alloc process borderlin pattern statist occurr frequenc close thresholdsoastohaveproperfrequentitemsetidentif featur call mine support count granular algorithm ram ds resourc awar mine data stream design reduc memori requir data storag retain good approxim target time seri experiment result shown memori requir store signific wavelet coeffici small qualiti approxim stabl perform increment data updat indic algorithm ram ds feasibl suitabl adapt mine data stream graph index frequent structur dbase approach graph increasingli import model complic structur schemaless data protein chemic compound xml document graph queri desir retriev graph quickli larg databas graph base indic paper investig issu index graph propos solut appli graph mine techniqu di erent exist path base method approach call gindex make frequent substructur basic index featur frequent substructur ideal candid explor intrins characterist data stabl databas updat reduc size index structur techniqu size increas support constraint discrimin fragment introduc perform studi show gindex time smaller index size achiev time perform comparison typic path base method graphgrep gindex approach eleg solut graph index problem demonstr databas index queri process benefit data mine frequent pattern mine concept develop appli index sequenc tree complic structur bass approxim search larg string databas paper studi problem build index structur larg string databas effici support type string match necess map substr numer space string tree mr index restrict memori practic suffix tree suffix arrai goal propos index scheme bass tree effici support gener approxim substr match term symbol substitut misalign sublinear time larg string databas kei idea design posit string group recurs fulli balanc tree similar subsequ segment start posit node label regular express describ common substr index subtre search properli direct portion databas high potenti match quickli bass tree place wild card queri pattern handl seamless addit search long pattern decompos seri search short segment process join result demonstr experi potenti perform improv brought bass tree order magnitud altern method fast algorithm subspac cluster pattern similar framework project cluster high dimension data stream data stream problem studi extens recent year great eas collect stream data natur stream data make essenti algorithm requir pass data recent singl scan stream analysi method propos context tempor dimens search web search singl import applic internet famou search techniqu pagerank hit algorithm algorithm motiv observ hyperlink page implicit convey author target page exploit social phenomenon identifi qualiti page author page hub page paper argu algorithm miss import dimens web tempor dimens web static environ constantli qualiti page past qualiti page futur techniqu favor older page page link accumul time page high qualiti inlink left bring qualiti page user import user latest inform research public search problem paper studi tempor dimens search context research public search propos number method deal problem experiment result show method highli effect mine surpris period pattern paper focu mine surpris period pattern sequenc event applic comput biologi infrequ pattern consid signific actual occurr frequenc exce prior expect larg margin tradit metric support necessarili ideal model measur kind surpris pattern treat pattern equal sens occurr carri weight assess signific pattern probabl occurr suitabl measur inform introduc natur degre surpris occurr pattern continu monoton decreas function probabl occurr pattern vastli occurr probabl handl seamlessli accumul degre surpris repetit pattern concept inform gain propos measur degre surpris pattern data sequenc bound inform gain properti identifi tackl predica caus violat downward closur properti inform gain measur turn effici solut problem empir test demonstr effici us propos model util comput sla manag base busi object editori state transact eic editori eic editori war weight associ rule item intens discov high dorder period pattern discoveri period pattern time seri data activ research area applic pattern hierarch natur higher level pattern consist repetit lower level pattern presenc nois prevent higher level pattern recogn sens portion data sequenc support high level pattern di erent layout occurr basic symbol exist common represent term raw symbol combin high level pattern express previou model defin raw symbol symbol combin properli recogn exist method paper propos model meta pattern captur high level pattern exibl model number potenti meta pattern larg substanti diculti li identifi proper pattern candid priori properti provid suffici prune power properti compon locat properti identifi conduct candid gener ecient comput base mine algorithm develop appli algorithm real synthet sequenc interest pattern discov enhanc biclust express data molecular biologi provid power tool express pattern thousand gene monitor simultan produc huge amount valuabl data concept biclust introduc cheng church captur coher subset gene subset condit set heurist algorithm design find biclust set biclust consist iter mask null valu discov biclust coars fine node delet node addit inclus invert data heurist inevit suffer drawback mask null valu discov biclust random number result phenomenon random interfer turn impact discoveri high qualiti biclust address issu acceler biclust process gener model biclust incorpor null valu propos probabilist algorithm floc discov set possibl overlap biclust simultan algorithm easili extend support addit featur suit requir virtual cost experiment studi yeast gene express data show floc algorithm offer substanti improv previous propos algorithm cross dfeatur analysi detect ad dhoc rout anomali algorithm content dbase public dsubscript system index weight dsequenc larg databas present index structur manag weight sequenc larg databas weight sequenc defin dimension structur element sequenc weight seri network event instanc weight sequenc event timestamp queri larg sequenc databas event occurr pattern step understand tempor causal relationship event index structur propos paper enabl effici retriev databas subsequ possibl contigu match queri sequenc event weight index method take consider uniform frequenc distribut event sequenc data addit method find broad rang applic index scientif data consist multipl numer column discoveri correl column instanc index dna micro arrai record express level gene condit enabl search gene respons experiment perturb follow pattern demonstr real world data set method effect effici build text classifi posit unlabel exampl paper studi problem build text classifi posit unlabel exampl kei featur problem neg learn recent techniqu solv problem propos literatur techniqu base idea build classifi step exist techniqu method step paper introduc method step perform comprehens evalu combin method step propos principl approach solv problem base bias formul svm show experiment accur exist techniqu mapl fast algorithm maxim pattern dbase cluster mine type arbitrarili posit overlap cluster genet data find interest pattern random model accuraci effici induct learn sequenti data scan manag ebusi demand sla contract busi term cross dsla execut manag sam imper competit busi servic provid posit manag execut servic level agreement sla contract busi term minim financi penalti servic level violat maxim servic level measur base custom satisfact metric paper briefli describ design rational integr set businessori servic level manag slm technolog develop sam project ibm watson research center busi sla execut manag sam enabl provid deploi effect mean captur manag contractu sla data provid face contractu slm data assist servic personnel priorit process action demand qualiti manag alert provid slm object autom priorit execut manag approv slm process behalf provid includ assign slm task servic personnel mine concept ddrift data stream ensembl classifi recent mine data stream concept drift action insight import challeng task wide rang applic includ credit card fraud protect target market network intrus detect convent knowledg discoveri tool face challeng overwhelm volum stream data concept drift paper propos gener framework mine concept drift data stream weight ensembl classifi train ensembl classif model ripper naiv bayesian sequenti chunk data stream classifi ensembl judici weight base expect classif accuraci test data time evolv environ ensembl approach improv effici learn model accuraci perform classif empir studi show propos method substanti advantag singl classifi approach predict accuraci ensembl framework effect varieti classif model appli data mine investig monei launder crime paper studi problem appli data mine facilit investig monei launder crime mlc identifi paradigm problem automat commun gener base uni parti data data direct explicit link inform propos methodolog link discoveri base correl analysi ldca mlc group model gener exemplari applic problem paradigm focus applic develop specif method automat mlc group model gener hierarch composit base hcb correl ldca methodolog prototyp call coral implement mlc group model gener applic preliminari test evalu base real mlc case data report contribut work identif uni parti data commun gener problem paradigm propos methodolog ldca solv problem paradigm formul mlc group model gener problem paradigm applic ldca methodolog develop specif solut mlc group model gener problem introduc hcb correl measur fuzzi logic develop evalu test coral prototyp real mlc case data categori subject descriptor stamp discoveri statist import pattern repeat long sequenti data paper focu mine period pattern allow degre imperfect form random replac perfect period pattern infomin propos metric gener inform gain identifi pattern event vastli occurr frequenc adjust deviat pattern penalti allow gap pattern occurr locat repeat dna sequenc paper present effect mine algorithm stamp simultan mine signific pattern subsequ model gener inform gain vist dynam index method queri xml data tree structur research provid flexibl queri facil extract data structur xml document paper propos vist index structur search xml document repres xml document xml queri structur encod sequenc show queri xml data equival find subsequ match unlik index method disassembl queri multipl queri join result queri provid final answer vist tree structur basic unit queri avoid expens join oper vist unifi index content structur xml document perform advantag method index content structur vist support dynam index updat reli sole tree special data structur support dbmss experi show vist ectiv scalabl cient support structur queri regress dbase tempor pattern mine scheme data stream devis paper regress base algorithm call algorithm ftp ds frequent tempor pattern data stream mine frequent tempor pattern data stream provid gener framework pattern frequenc count algorithm ftp ds major featur data scan onlin statist collect regressionbas compact pattern represent attain featur data scan data segment pattern growth scenario explor frequenc count purpos algorithm ftp ds scan onlin transact flow gener candid frequent pattern real time import featur algorithm ftp ds regress base compact pattern represent specif meet space constraint devis pattern represent compact atf stand accumul time frequenc form aggreg compris inform requir regress analysi addit develop techniqu segment tune segment relax enhanc function ftp ds featur algorithm ftp ds conduct mine variabl time interv perform trend detect effect synthet data real dataset net cooper transcach system distribut proxi server web content adapt score data associ rule data mine applic object select data case target class guest editori text web mine replic load balanc hot dspot relief proxi web cach hash rout editori ae introduct optim index alloc sequenti data broadcast wireless mobil comput mine asynchron period pattern time seri data periodici detect time seri data challeng problem great import applic editori ae introduct request redirect algorithm distribut web system data mine research meet practic develop ieee intern confer data mine san jose california novemb decemb panel discuss data mine research meet practic develop motiv organ panel discuss provid advic industri peopl explor direct data mine develop advanc knowledg discoveri data mine prune dynam schedul cost dsensit ensembl effici queri monitor adapt multipl kei hash index structur pattern similar search dna microarrai dat acceler approxim subsequ search larg protein sequenc databas index structur pattern similar search dna microarrai dat cluster algorithm content dbase public dsubscript system effici commun scheme base network support applic level multicast techniqu content base public subscript system show commun cost depend heavili network configur distribut public subscript devis algorithm adapt exist partit data cluster algorithm algorithm determin multicast group common base total subscrib interest perform context highli heterogen subscript scale effici respect ideal solut achiev small number multicast group experi concept appli match public subscrib real time determin dynam unicast multicast broadcast inform event network match subscrib demonstr qualiti algorithm simul experi fulli distribut framework cost dsensit data mine paper propos fulli distribut system compar central partial distribut system cost sensit data mine experiment result shown approach achiev higher accuraci central partial distribut learn method incur train time commun comput overhead delta dcluster captur subspac correl larg data set cluster activ research area great practic import recent year previou cluster model focus group object similar valu set dimens subspac cluster assum object dimens biclust exist cluster model adequ captur coher exhibit object strong coher exist set object subset attribut valu attribut attribut valu fulli common applic includ bio informat analysi collabor filter analysi data incomplet subject bias bio informat biclust model recent propos captur coher subset attribut introduc gener model refer ffi cluster model captur coher exhibit subset object subset attribut allow absent attribut valu move base algorithm floc devis effici produc optim cluster result ffi cluster model take biclust model special case floc algorithm perform superior biclust algorithm demonstr correct effici ffi cluster model floc algorithm number real synthet data set progress model present induct learn perform frustrat batch process user interact system control final accuraci train time accuraci produc model low comput resourc misspent paper propos progress model framework progress model learn algorithm estim onlin accuraci final model remain train time estim accuraci expect user termin train prior complet wast resourc user choos complet learn process progress model comput model expect accuraci expect time describ implement progress model ensembl classifi mine associ pattern structur larg relat tabl associ rule mine aim discov pattern support threshold mine pattern compos item arbitrari subset attribut larg relat tabl repres challeng practic applic includ event manag system motiv work attribut combin defin item pattern provid structur inform pattern current associ algorithm make full structur inform pattern inform lost encod attribut valu constrain hierarchi taxonomi pattern structur convei import knowledg pattern paper present architectur organ mine space base pattern structur exploit interrelationship pattern structur execut time mine reduc significantli advantag demonstr experi synthet real life dataset infomin mine partial period pattern gap penalti paper focu mine period pattern allow degre imperfect form random replac perfect period pattern inform gain propos identifi pattern event vastli occurr frequenc adjust deviat pattern penalti exist gap pattern occurr applic bio informat import identifi subsequ pattern repeat perfectli perfectli solut extend inform gain measur includ penalti gap pattern occurr call measur gener inform gain find subsequ pattern gener inform gain high locat repeat dna sequenc paper develop effect mine algorithm infomin simultan mine signific pattern subsequ partial supervis classif text document investig problem set document topic class larg set mix document document class type document identifi document class kei featur problem label document make tradit machin learn techniqu inapplic label document class call problem partial supervis classif paper show problem pose constrain optim problem condit solut constrain optim problem give good solut partial supervis classif problem present techniqu solv problem demonstr effect techniqu extens experiment distribut data mine chain store databas short transact paper broaden horizon tradit rule mine introduc framework causal rule mine distribut chain store databas specif causal rule explor paper consist sequenc trigger event set consequenti event design capabl mine sequenti intertransact inform causal rule mine gener framework rule deriv note procedur causal rule mine costli presenc huge number candid set distribut databas opinion dealt direct extens exist rule mine method devis paper seri level match algorithm includ level match abbreviat lm level match select scan abbreviatedli lm distribut level match abbreviatedli distibut lm minim comput cost need distribut data mine causal rule addit phenomena time window constraint consider develop algorithm result properli emploi technolog level match select scan propos algorithm present good effici scalabl mine local global causal rule scale experi show propos algorithm scale action effici filter larg dataseta user dcentric paradigm mine relationship trigger consequenti event short transact databas cluster pattern similar larg data set cluster process group set object class similar object definit similar vari cluster model model concept similar base distanc euclidean distanc cosin distanc word similar object requir close valu set dimens paper explor gener type similar pcluster model propos object similar exhibit coher pattern subset dimens instanc dna microarrai analysi express level gene rise fall synchron respons set environment stimuli magnitud express level close pattern exhibit alik discoveri cluster gene essenti reveal signific connect gene regulatori network commerc applic collabor filter benefit model captur close valu lead indic close purchas brows pattern exhibit custom paper introduc effect algorithm detect cluster perform test real synthet data set show effect mine long sequenti pattern noisi environ applic includ comput biologi studi consum behavior analysi system perform analysi noisi environ observ sequenc accur reflect underli behavior protein sequenc amino acid mutat impact biolog function protein desir occurr observ relat mutat manner support measur number occurr pattern serv purpos paper introduc concept compat matrix mean provid probabilist connect observ underli true metric match propos captur real support pattern expect nois free environ assum addit context address pattern long standard prune techniqu develop market basket problem work effici result algorithm combin statist sampl techniqu border collaps devis discov long pattern minim number scan sequenc databas suffici high confid empir result demonstr robust match model respect nois effici probabilist algorithm autom system web portal person paper propos system person web portal specif implement discuss refer web portal new feed servic techniqu propos effect categor manag person new feed obtain live new wire servic process consist step manual input requir build domain knowledg site specif autom compon domain knowledg order perform person categor present effect scheme advertis propos target inform user content web page advertis icon appear autom techniqu identifi sudden variat new pattern support new alert descript version softwar custom web site provid dynam refin tabl summar dcommerc optim crawl strategi web search engin control replic hash rout dbase web cach state art local distribut web dserver system control replic hash rout dbase web cach find local associ market basket data paper discuss techniqu discov local associ segment data cluster aggreg behavior data set di erent local segment case desir design algorithm ectiv discov local associ expos custom pattern speci aggreg behavior inform beveri target market present empir result show method signi cantli larger number associ discov analysi aggreg data editori call amp calendareditori redefin cluster high ddimension applic high dimension data challeng cluster algorithm inher sparsiti point recent research result high dimension data concept proxim cluster meaning discuss gener techniqu project cluster construct cluster arbitrarili align subspac lower dimension subspac specif cluster definit substanti gener realist techniqu limit method project origin set attribut gener project cluster techniqu view redefin cluster high dimension applic search hidden subspac cluster creat inter attribut correl provid concept extend cluster featur vector order make algorithm scalabl larg databas run time space requir algorithm adjust tradeoff accuraci perform studi robust load share strategi distribut heterogen web server system replic inform multipl server common approach support popular web site distribut architectur mechan assign client request web server scalabl central mirror architectur paper distribut system authorit domain server adn web site take request dispatch role map url hostnam ip address visibl node web server web cluster interfac architectur support local geograph distribut web server adn control small fraction request reach web site address map request client access reduc internet trac address resolut cach server time live ttl period open set problem tradit central schedul parallel distribut system face heterogen assumpt web node capac practic increas order complex request assign problem ieee ieee transact knowledg data engin vol pp march april person materi permit permiss reprint republish materi advertis promot purpos creat collect work resal redistribut server list reus copyright compon work work obtain ieee contact manag copyright permiss ieee servic center hoe lane box piscatawai nj usa telephon intl fast algorithm onlin gener profil associ rule introduc ae adapt piggyback scheme video don ddemand system effect nearest neighbor index euclidean metric nearest neighbor search import oper wide multimedia databas higher dimens previou method nearest neighbor search ineffici requir comput nearest neighbor distanc larg fraction point space paper present approach process nearest neighbor search euclidean metric search small subset origin space approach effect approxim cluster encapsul geometr regular shape comput upper lower bound distanc queri point cluster show effect propos approach perform extens experi result reveal propos approach significantli outperform tree sequenti scan effect conceptu index similar search text data similar search text proven interest problem qualit perspect inher redund ambigu textual descript method search engin order retriev document similar user defin set keyword applic target medium larg size document greater nois effect stem presenc larg number word unrel topic document invert represent domin method index text suitabl document document similar search short user queri improv qualiti similar search latent semant index lsi map document origin set word concept space lsi map data domain provid effect index techniqu paper investig wai provid conceptu search document creat represent term conceptu word chain techniqu effect index techniqu similar queri perform larg collect document access small amount data demonstr scheme outperform standard textual similar search invert represent term qualiti search effici ssdt scalabl subspac dsplit classifi bias data meta dpattern reveal hidden period pattern discoveri period pattern time seri data activ research area applic pattern hierarch natur higher level pattern consist repetit lower level pattern presenc nois prevent higher level pattern recogn sens portion data sequenc support high level pattern layout occurr basic symbol exist common represent term raw symbol combin high level pattern express previou model defin raw symbol symbol combin properli recogn exist method paper propos model meta pattern captur high level pattern flexibl model number potenti meta pattern larg substanti difficulti li identifi proper pattern candid apriori properti provid suffici prune power properti compon locat properti identifi conduct candid gener effici comput base mine algorithm develop appli algorithm real synthet sequenc interest pattern discov discov unexpect inform competitor web site begin web find inform web import problem exist approach includ keyword base search wrapper base inform extract web queri user prefer approach essenti find inform match user explicit specif paper argu insuffici type inform great interest unexpect inform unanticip user find unexpect inform applic compani find unexpect inform competitor unexpect servic product competitor offer inform compani learn competitor design counter measur improv competit number page typic commerci site larg relev site competitor difficult human user view page discov unexpect inform autom assist need paper propos number method user find type unexpect inform competitor web site experi result show techniqu practic effici keyword inform interesting web comparison web mine infomin mine surpris period pattern lessoftheprobabilityofoccurr amoresuitablemeathesensethateveryoccurrencecarriesthesameweightto server theworkloadoneachserverismeasur wardstheassessmentofthesignicanceofapatternregard rank low relativelylow relativelyhigh andhigh thentherear dierentev oneforeach surement inform isintroducedtonaturallyvalueth possiblecombinationofserverst somepreliminari uousandmonotonicallydecreasingfunctionofitsprobabilitydegreeofsurpriseofeachoccurrenceofapatternasacontin ofoccurr thiswouldallowpatternswithvastlydier occurrenceprobabilitiestobehandledseamlessli astheacoccuratamuchlowerfrequencythanotherst pat examinationoftheclusterstatesovertimemightshow thatthestateuctuationcomplieswithsomeperiod cumulateddegreeofsurpriseofallrepetitionsofapattern thealldegreeofsurpriseofthepatternwithinadatasequ conceptofinformationgainisproposedtomeasuretheovertemadministratorsiftheoccurrenceofsuchpattern tern involvingitmaybeofmoreintereststosomesi behavior althoughthehighworkloadonallserversmai theboundedinformationgainpropertyisidentiedtotackl thepredicamentcausedbytheviolationofthedownwardclo earthquak earthquakesoccurveryoftenincalifornia surepropertybytheinformationgainmeasureandinturn demonstratetheeciencyandtheusefulnessofthepropos providesanecientsolutiontothisproblem empiricaltest maybeinterestedinknowingwhetherthereexistsani contradictsthepriorexpect inherentseismicperiodsothatpredictioncanbemad itcanbeclassiedbyitsmagnitudeandtyp scientist model outlier detect high dimension data outlier detect problem import applic eld fraud detect netw ork robust analysi intrus detect suc applic high dimension domain whic hthe data con tain hundr dimens recen algorithm concept pro ximiti order outlier base relationship rest data ho high dimension space data spars notion proxim fail retain meaning fact sparsiti high dimension data impli point equal good outlier perspect ofproxim base de nition high dimension data notion nding meaning outlier substanti complex obviou paper discuss techniqu outlier detect whic outlier studi behavior project data set segment dbase proxi cach multimedia stream stream video audio internet popular proper proxi cach larg multimedia object increasingli import larg media object hour video treat video singl web object cach paper present evalu segment base buffer manag approach proxi cach larg media stream block media stream receiv proxi server group variabl size segment cach admiss replac polici attach cach valu segment take account segment distanc start media cach polici give preferenti treatment begin segment user quickli plai back media object delai event driven simul conduct evalu segment base proxi cach approach result show segment base cach effect increas byte hit ratio reduc total traffic lower number request requir delai start segment base cach advantag cach size limit set hot media object time media file size larg user stop plai media initi block intellig crawl world wide web arbitrari predic enorm growth world wide web recent year made import perform resourc discoveri effici idea propos recent year kei techniqu focus crawl crawl topic portion world wide web quickli explor web page paper propos concept intellig crawl learn characterist linkag structur world wide web perform crawl specif intellig crawler inlink web page content candid url structur behavior inlink web page sibl order estim probabl candid crawl gener framework focus crawl techniqu base pre defin understand topic structur web techniqu discuss paper applic crawl web page satisfi arbitrari user defin predic topic queri keyword queri combin unlik focus crawl provid repres topic exampl crawler learn topic refer techniqu intellig crawl adapt natur adjust web page linkag structur learn crawler capabl reus knowledg gain crawl order provid effici crawl close relat predic design learn crawler topic resourc discoveri balanc load cluster web farm thrash control avoid concurr mergesort parallel prefetch maximum factor queue length batch scheme video don ddemand system editori call amp calendareditori editori introduc ae approach onlin gener associ rule schedul algorithm broadcast deliveri digit product mine associ collect strength approach load balanc cluster web farm cluster decis tree construct paper propos cluster techniqu base supervis learn techniqu call decis tree construct techniqu overcom shortcom kei idea decis tree partit data space cluster empti spars region level detail techniqu find natur cluster larg high dimension space effici suitabl cluster full dimension space subspac comprehens descript cluster experi result synthet data real life data show techniqu effect scale larg high dimension dataset collabor proxi system distribut web content transcod tabsum flexibl dynam tabl summar approach intellig recommend analyz igrid index revers dimension curs similar index high dimension space similar searc index problem ell kno wn di cult high dimension applic index structur show rapid degrad increas dimension whic hlead access en tire databas queri urthermor recen research result sho high dimension space concept similar meaning paper propos theigrid index method similar index distanc function meaning retain increas dimension addit techniqu show perform uniqu index structur percentag data access invers proport data dimension techniqu reli dimension high order pro vide perform cient similar result igridindex support special kind queri whic hw refer project rang queri queri whic increasingli relevan tfor high dimension data mine applic effici mine weight associ rule war paper extend tradit associ rule problem allow weight item transact reflect interest intens item transact turn opportun associ weight paramet item result associ rule call weight associ rule war war improv confid rule mechan effect target market identifi segment custom base potenti degre loyalti volum purchas approach mine war ignor weight find frequent itemset tradit frequent itemset discoveri algorithm introduc weight rule gener shown experiment result approach result shorter averag execut time produc higher qualiti result gener previou method quantit associ rule mine asynchron period pattern time seri data periodici detect time seri data challeng problem great import applic target student data mine educ domain offer fertil ground interest challeng data mine applic applic educ student improv qualiti educ paper present real life applic gift educ programm gep ministri educ moe singapor applic involv data mine task paper focus task select student remedi class tradition cut mark subject select weak student student score subject fall cut mark subject advis class subject paper show tradit method requir student part remedi class increas teach load teacher unnecessari burden student undesir case gep student gener take subject gep student gep student encourag time explor advanc topic data mine select target student precis geograph load balanc scalabl distribut web system user highli popular web site experi long delai access inform upgrad content site infrastructur singl node local distribut web cluster compos multipl server node limit relief cluster wide area connect bottleneck solut distribut web cluster internet place content node strateg locat geograph distribut architectur domain system dn server evalu network proxim user serv closest cluster reduc network impact respons time hand serv closest request unbalanc server increas system impact respons time achiev scalabl web system propos integr dn proxim schedul http request redirect mechan web server activ demonstr simul experi dispatch mechan augment percentag request guarante respons time enhanc qualiti servic geograph distribut web site http request redirect select addit round trip increas network impact latenc time experienc user contribut paper propos compar mechan limit reassign neg consequ load balanc find gener project cluster high dimension space high dimension data challeng cluster algorithm inher sparsiti point recent research result high dimension data concept proxim cluster meaning discuss gener techniqu project cluster construct cluster arbitrarili align subspac lower dimension subspac specif cluster definit substanti gener realist techniqu limit method project origin set attribut gener project cluster techniqu view redefin cluster high dimension applic search hidden subspac cluster creat inter attribut correl provid concept extend cluster featur vector order make algorithm scalabl larg databas run time space requir algorithm adjust tradeoff accuraci latenc dsensit hash collabor web cach workfil disk manag concurr mergesort multiprocessor databas system data mine techniqu person paper discuss overview data mine techniqu person discuss standard techniqu order adapt increas abil system tailor specif user behavior discuss techniqu collabor filter content base method content base collabor filter method examin specif applic techniqu scenario broad advantag specif situat mine pattern long sequenti data nois report intern workshop advanc issu dcommerc web dbase inform system local replic proxi web cach hash rout data mine person technolog redirect algorithm load share distribut web dserver system support high request rate popular web site cluster web server organ prefer multipl independ mirror server maintain singl interfac user potenti scalabl fault toler load balanc paper propos web cluster architectur domain system server dn dispatch user request server url ip address map mechan integr redirect request mechan base http protocol allevi side effect cach ip address map intermedi server compar altern mechan includ synchron asynchron activ central distribut decis redirect analyz reassign entir domain individu client request type statu inform server select polici redirect request result show combin central distribut dispatch polici web server cluster handl high load skew www environ load balanc hot spot relief hash rout collect proxi cach hort hatch egg graph dtheoret approach collabor filter paper introduc approach ratingbas collabor filter techniqu commerc merchant offer group homogen item compact disk video book softwar contrast collabor filter techniqu algorithm graph theoret base twin concept hort predict demonstr paper techniqu fast scalabl accur requir modest learn curv make hierarch classif scheme order introduc context rate process call creativ link order find surpris atyp item recommend item cross group boundari techniqu kei engin intellig recommend algorithm ira project develop ibm research addit recommend engin ira situat analyz determin mix engin commerc merchant engin optim placement advertis adapt queri process time dseri data ywh ibm tradit queri process time seri data transform data time domain frequenc domain control end user isn suit queri find pattern dynam user constraint queri present method search time seri data transform time sequenc symbol string chang ratio contigu data point time seri suffix tree built index suffix symbol string focu paper demonstr method adapt process dynam constrain time seri queri merit build categor system supervis cluster paper investig supervis cluster order creat set categori classif document inform pre exist taxonomi order supervis creation set relat cluster freedom defin creat class show advantag supervis cluster control rang subject categor system address precis mathemat definit categori categor document priori knowledg definit categori discuss techniqu classifi distinguish close relat cluster final show empir categor system util machin deriv taxonomi perform manual categor process lower cost optim profit broadcast deliveri multimedia product data mine techniqu associ cluster classif replic issu proxi cach arrai text mine techniqu person method similar index market basket data recent year data mine method propos find structur inform market basket data associ rule model recent propos order discov pattern depend data paper discuss method index market basket data effici similar search techniqu applic util similar custom bui behavior order make peer recommend propos index call signatur tabl flexibl support wide rang similar function construct index structur independ similar function queri time result similar search algorithm show excel scalabl increas memori avail databas size fast algorithm project cluster research text cluster applic text cluster exist text cluster system review mine associ rule set item larg databas dynam load balanc web dserver system cach world wide web continu increas demand inform servic internet show sign strain internet highli distribut system individu data object singl sourc host comput network link easili overload larg number user access popular data proxi cach popular reduc network bandwidth server load improv respons time user origin cach proxi cern wide paper describ softwar develop author investig altern techniqu cach world wide web object softwar complement tradit proxycach allow server explicitli grant deni permiss cach object support server initi callback invalid chang object introduct statist collect nsfnet backbon indic world wide web traffic exce scan hierarch algorithm similar search databas consist long sequenc unbalanc tree index multidimension object run placement polici concurr mergesort parallel prefetch design analysi permut dbase pyramid broadcast dn dispatch algorithm state estim scalabl web dserver cluster replic inform server cluster promis support popular web site web server cluster requir mechan schedul request server common approach local domain server dn central dispatch main problem www address cach mechan reduc network trac dn dispatch control small fraction request reach web server cluster nonuniform load dierent client domain high variabl real web workload introduc addit degre complex load balanc issu characterist make exist schedul algorithm tradit distribut system applic control load web server cluster motiv research dn polici requir system state inform analyz dn dispatch polici realist situat state inform estim low comput commun overhead applic web cluster architectur model realist scenario web cluster larg set simul experi show incorpor propos state estim dispatch polici eectiv dn schedul algorithm improv substanti compar result dn algorithm adequ state inform malm framework mine sequenc databas multipl abstract level malm framework mine sequenc databas multipl abstract level onlin algorithm find profil associ rule effici state estim load control polici scalabl web server cluster replic inform server cluster promis support popular web site web server cluster requir mechan direct request server common approach domain server dn central schedul address cach mechan uniform load client domain complic load balanc issu make exist schedul algorithm tradit distribut system applic web server cluster paper theoret dn polici requir system state inform extend realist situat state inform estim low comput commun overhead show incorpor estim dn polici load balanc improv substanti dn control limit small portion client request rang dbase bitmap index high cardin attribut skew dynam load balanc geograph distribut heterogen web server increas web traffic distribut multi server web site provid scalabl flexibl cope grow client demand load balanc algorithm spread request multipl web server crucial achiev scalabl domain server dn base schedul propos literatur multipl homogen server presenc heterogen web server increas complex dn schedul problem make previous propos algorithm homogen distribut system directli applic lead propos polici call adapt ttl algorithm account uneven distribut client request rate heterogen web server adapt set time live ttl address map request extens simul result show strategi robust effect balanc load geograph distribut bandwidth dsensit updat schedul method internet push onlin gener associ rule larg databas consist sale transact investig problem onlin mine associ rule larg databas show preprocess data effect order make suitabl repeat onlin queri preprocess algorithm take account storag space store preprocess data onlin process appli graph theoret search algorithm complex proport size output result onlin algorithm practic instantan term respons time algorithm support techniqu quickli discov associ rule larg itemset algorithm capabl find rule specif item anteced consequ associ rule present compact form elimin redund elimin redund onlin gener associ rule larg itemset interest onlin gener profil associ rule framework itemset gener problem find associ rule larg databas sale transact wide studi literatur discuss weak larg itemset method associ rule gener method evalu find itemset refer strongli collect itemset propos concept support itemset correl item itemset relat criterion stress import actual correl item absolut support previous propos method provid correl itemset necessarili applic larg databas provid algorithm good comput effici maintain statist robust fact algorithm reli rel measur absolut measur support impli method appli find associ rule dataset item sizeabl percentag transact dens dataset dataset item vari densiti neg associ rule framework optim www advertis perform studi collabor method hierarch cach proxi server energi deffici mobil cach invalid cach data wireless mobil comput significantli reduc bandwidth requir due batteri power limit wireless mobil comput forc oper doze total disconnect mode result mobil comput miss cach invalid report paper present energi effici cach invalid method wireless mobil comput cach invalid scheme call group cold updat set retent gcore wake mobil comput check cach valid server reduc bandwidth requir valid check data object partit group simpli invalid group object group updat gcore retain cold updat set object group present effici implement gcore conduct simul evalu cach effect result show gcore sub speedtrac web usag mine analysi tool usag mine analysi tool speedtrac world wide web usag mine analysi tool develop understand user surf behavior explor web server log file data mine techniqu popular web explod strong desir understand user surf behavior difficult perform userori data mine analysi directli server log file tend ambigu incomplet innov algorithm speedtrac identifi user session reconstruct user travers path requir cooki user registr session identif user privaci protect user session identifi data mine algorithm appli discov common travers path group page frequent visit import user brows pattern manifest frequent travers path page group help understand user surf behavior type report prepar user base report path base report group base report paper describ design speedtrac demonstr featur sampl report opular world wide web www internet explod recent organ invest tremend amount capit oper site web web site provid commun servic employe custom supplier monei invest site strong desir understand effect invest find wai realiz potenti opportun mine larg itemset associ rule paper survei itemset method associ rule gener paper discuss past research topic studi relev import itemset method gener associ rule discuss number variat associ rule problem propos literatur practic applic inher weak larg itemset method associ rule gener explor discuss formul associ viabl altern tradit associ rule gener method introduct associ rule find relationship item databas sale transact rule track bui pattern consum behavior find presenc item transact affect presenc problem associ rule gener recent gain consider promin analysi task assign polici scalabl distribut web dserver system distribut multi server web site provid scalabl grow client demand popular site load balanc distribut web server system consist multipl web server document retriev domain server dn address resolut open interest problem paper investig effect activ dn atyp central schedul appli schedul strategi rout request suitabl web server unlik tradit parallel distribut system central schedul full control system dn control small fraction request reach multi server web site peculiar presenc highli skew load make difficult achiev accept load balanc avoid overload web server paper adapt tradit schedul algorithm dn propos polici effici data mine path travers pattern abstract paper explor data mine capabl involv mine path travers pattern distribut inform provid environ document object link facilit interact access solut procedur consist step deriv algorithm convert origin sequenc log data set maxim forward refer filter effect backward refer made eas travel concentr mine meaning user access sequenc deriv algorithm determin frequent travers pattern larg refer sequenc maxim forward refer obtain algorithm devis determin larg refer sequenc base hash prune techniqu improv option determin larg refer sequenc batch reduc number databas scan requir perform method compar analyz shown option select scan advantag lead promin perform improv sensit analysi paramet conduct index term data mine travers pattern distribut inform system world wide web perform analysi analysi task assign polici scalabl distribut web dserver system distribut multi server web site provid scalabl grow client demand popular site load balanc distribut web server system consist multipl web server document retriev domain server dn address resolut open interest problem paper investig effect activ dn atyp central schedul appli schedul strategi rout request suitabl web server unlik tradit parallel distribut system central schedul full control system dn control small fraction request reach multi server web site peculiar presenc highli skew load make difficult achiev accept load balanc avoid overload web server paper adapt tradit schedul algorithm dn propos polici explor wait toler effect batch video don ddemand schedul video demand vod comput system batch request video share common data stream lead signific improv throughput wait toler characterist commonli observ viewer behavior introduc paradigm schedul vod system propos analyz class schedul scheme max batch min idl scheme provid altern wai stream capac effect batch make video select propos scheme consider stream complet time viewer wait toler compar propos scheme previous studi scheme firstserv fcf schedul video longest wait request maximum queue length mql scheme select video maximum number wait request show simul propos scheme outperform substanti fc increas multimedia system throughput consumpt dbase buffer manag analyt model multimedia batch scheme smart smart bound weight respons time schedul system independ task schedul preemption parallel comput task number processor requir execut time weight problem schedul minimum weight averag respons time present algorithm call smart problem produc solut factor optim knowledg rst polynomi time algorithm minimum weight averag respons time problem achiev constant bound addit unweight case weight uniti describ variant smart produc solut factor optim improv bound special case reminisc influenti paper mine associ rule adjust accuraci paper devis effici algorithm mine associ rule adjust accuraci note applic requir mine transact data captur custom behavior frequent applic effici data mine import factor requir complet accuraci mine result allow imprecis result significantli improv data mine effici paper method mine associ rule adjust accuraci develop deal concept sampl method obtain essenti knowledg sampl subset light knowledg perform effici associ rule mine entir databas techniqu relax support factor base sampl size devis achiev desir level accuraci method differ wai util sampl data perform meth disk cach web object proxi server disk cach web object proxi server index sequenti data broadcast wireless mobil comput energi save import issu wireless mobil comput viabl approach achiev energi save index data organ broadcast data wireless channel mobil unit paper explor issu index data skew access sequenti broadcast wireless mobil comput propos method build index tree base access frequenc data record minim averag cost index probe case fix index fanout variant index fanout devis algorithm construct index tree case show cost index probe minim emploi imbalanc index tree design accord data access skew exploit variant fanout index node introduct mobil comput refer activ person digit assist palmtop notebook compu schedul algorithm distribut web server cluster architectur parallel sysplex scalabl diverg control algorithm epsilon serializ paper present distribut diverg control algorithm epsilon serializ homogen heterogen distribut databas epsilon serializ concurr permit serializ interleav databas oper epsilon transact present strict phase lock diverg control algorithm optimist diverg control algorithm homogen distribut databas system local order transact distribut epsilon transact environ total inconsist distribut epsilon transact simpli sum transact describ diverg control algorithm heterogen distribut databas system local order transact distribut epsilon transact total inconsist distribut epsilon transact greater optim design multipl hash tabl concurr control hash dbase method transact trim mine associ rule recoveri analysi data share system defer dirti page propag polici parallel execut hash join parallel databas paper explor import issu processor alloc hash filter improv parallel execut hash join exploit opportun pipelin hash join execut scheme transform bushi execut tree alloc tree devis alloc tree node denot pipelin concept synchron execut time processor alloc node alloc tree relat pipelin made approxim time approach hash filter investig improv parallel execut hash join extens perform studi conduct simul demonstr import processor alloc evalu scheme hash filter experiment shown processor alloc gener domin factor perform effect hash filter disk load balanc video don ddemand system video demand comput system propos scheme balanc load disk help solv perform problem crucial achiev maxim video throughput load balanc scheme consist compon static compon determin good assign video group stripe disk dynam compon assign featur dasd danc algorithm perform realtim disk schedul effect manner scheme work synergist disk stripe examin perform propos algorithm simul experi introduct video demand vod comput system consist central processor collect share disk direct access storag devic dasd vod comput system plai multipl stream video simultan base custom demand video store cost effect disk rotat mirror declust replica placement disk darrai dbase video server video demand vod environ disk arrai support disk bandwidth requir pose problem disk bandwidth disk failur paper explor approach replic frequent access movi provid high data bandwidth fault toler requir disk arrai base video server isochron continu video stream impos requir random access pattern databas file explicitli propos replica placement method call rotat mirror declust rmd support high data avail disk arrai vod environ essenc rmd similar convent mirror declust replica store disk arrai replica placement disk arrai rmd properli rotat combin merit prior chain mirror declust method rmd adapt ttl scheme load balanc distribut web server appli hash filter improv execut multi djoin queri lt gt paper explor approach interleav bushi execut tree hash filter improv execut multi join queri similar semi join distribut queri process hash filter appli elimin match tupl join relat execut join reduc join cost note hash filter built execut stage bushi tree cost effect effect hash filter evalu effici scheme determin effect sequenc hash filter bushi execut tree develop hash filter built appli base join sequenc bushi tree reduct effect optim cost minim scheme hash filter implement evalu simul experiment shown applic hash filter gener power effici data mine path travers pattern distribut system energi deffici cach wireless mobil comput hierarchyscan hierarch similar search algorithm databas long sequenc permut dbase pyramid broadcast scheme video don ddemand system consumpt dbase buffer manag maxim system throughput multimedia system optim batch polici video don ddemand storag server optim piggyback merg polici video don ddemand system abstract critic issu perform video demand system bandwidth quir order satisfi client request number techniqu propos order reduc bandwidth requir paper concentr techniqu adapt piggyback ing develop analyz piggyback merg poli ci optim larg class reason method model analysi concurr control scheme analyt model cluster raid map base random permut coupl multipl system global buffer paper conduct perform studi coupl multipl system global buffer present result obtain multipl system simul simul run workload coupl system behavior input studi statist includ local global buffer hit page write global buffer cross invalid castout relationship degre data skew explor addit updat cach approach design altern global buffer read cach explor read cach updat page page read node global buffer facilit node access page cost higher global buffer usag investig case cach global buffer simul result present analyz index term da optim parallel execut multi djoin queri paper studi subject exploit inter oper parallel optim execut multi join queri specif focu major issu schedul execut sequenc multipl join queri ii determin number processor alloc execut join oper obtain issu propos evalu simul method determin gener join sequenc bushi tree simplic heurist propos lead gener join sequenc significantli outperform optim sequenti join sequenc qualiti join sequenc obtain propos heurist shown fairli close optim issu shown processor alloc exploit inter oper parallel subject constraint execut depend system fragment studi data mine overview databas perspect perform analysi dynam finit version scheme storag cost obsolesc effici lru dbase buffer lan remot cach architectur gener result dto dall broadcast broadcast refer process node broadcast piec inform node system paper develop broadcast scheme deal class scheme prior scheme base gener minim complet set scheme base propag expert develop complet broadcast minim number step design minim number messag perform class scheme compar analyz broadcast scheme desir deriv combin advantag class scheme index term broadcast messag pass nodup expert propag commun step introduct share memori global clock synchron commun process node distribut system messag pass content dbase search download digit video client station proceed eleventh intern confer data engin rotat mirror declust replica placement disk darrai dbase video server video demand vod environ disk arrai support disk bandwidth requir pose problem disk bandwidth disk failur paper explor approach replic frequent access movi provid high data bandwidth fault toler requir disk arrai base video server isochron continu video stream impos requir random access pattern databas file explicitli propos replica placement method call rotat mirror declust rmd support high data avail disk arrai vod environ essenc rmd similar convent mirror declust replica store disk arrai replica placement disk arrai rmd properli rotat combin merit prior chain mirror declust method rmd effici parallel data mine associ rule paper develop algorithm call pdm conduct parallel data mine associ rule transact collect item larg itemset set item number transact exce pre threshold pdm design global set larg itemset identifi effici amount inter node data exchang requir minim specif databas partit process node collect count inform itemset local databas effici hash method inform discov node share node commun scheme pdm emploi techniqu call clue andpol address uncertainti due partial knowledg collect node judici select small fraction itemset exchang count inform node reduc commun cost global effici parallel data mine associ rule paper develop algorithm call pdm conduct parallel data mine associ rule transact collect item larg itemset set item number transact exce pre threshold pdm design global set larg itemset identifi effici amount inter node data exchang requir minim specif databas partit process node collect count inform itemset local databas effici hash method inform discov node share node commun scheme pdm emploi techniqu call clue andpol address uncertainti due partial knowledg collect node judici select small fraction itemset exchang count inform node reduc commun cost global perform model design client fserver system perform model design client fserver system dasd danc disk load balanc optim scheme video don ddemand comput effect hash base algorithm mine associ rule paper examin issu mine associ rule item larg databas sale transact mine associ rule map problem discov larg itemset larg itemset group item suffici number transact problem discov larg itemset solv construct candid set itemset identifi candid set itemset meet larg itemset requir gener iter larg itemset increas order larg itemset larg itemset item determin larg itemset huge number candid larg itemset earli iter domin factor data mine perform address issu propos effect hash base algorithm candid set gener explicitli number candid itemset gener propos algorithm order magnitud smaller previou method resolv perform bottleneck note gener smaller candid set enabl effect trim transact databas size earlier stage iter reduc comput cost iter significantli extens simul studi conduct evalu perform propos algorithm perform studi workfil disk manag concurr mergesort multiprocessor databas system diverg control distribut databas system paper present distribut diverg control algorithm epsilon serializ homogen heterogen distribut databas epsilon serializ concurr permit serializ interleav databas oper epsilon transact present strict phase lock diverg control algorithm optimist diverg control algorithm homogen distribut databas system local order transact distribut epsilon transact environ total inconsist distribut epsilon transact simpli sum transact describ diverg control algorithm heterogen distribut databas system local order transact distribut epsilon transact total inconsist distribut epsilon transact greater analyt model comparison stripe strategi replic disk arrai appli segment ddeep tree pipelin multipl hash join pipelin execut multijoin queri multiprocessor base databas system explor paper hash base join multipl join pipelin earli result join join complet join process execut queri denot queri execut tree improv execu tion pipelin hash join innov approach queri execut tree select propos exploit segment deep tree bushi tree deep subtre deriv analyt mode execut pipelin segment light model develop heurist scheme de termin queri execut plan base segment deep tree queri effici execut shown simul propos approach incur addit overhead plan execut possess flexibl queri plan gener lead queri plan perform achiev previou scheme deep tree hierarch approach parallel multiqueri schedul storag retriev method support fulli interact playout disk darrai dbase video server import challeng video demand vod system support interact brows function fast forward fast backward typic function impos addit resourc requir vod system term storag space retriev throughput network bandwidth preval video compress techniqu mpeg impos addit constraint process introduc inter frame depend paper devis method support variabl rate brows mpeg video stream minim addit resourc requir specif storag retriev video data disk arrai base video server address issu distribut retriev request disk evenli approach propos paper interact brows compos storag method sampl placement method playout method sampl plac design analysi dahead schedul scheme support paus dresum video don ddemand applic video demand vod system subscrib choos movi view time view environ invari hot video request viewer requir viewer independ paus video instant resum view delai difficulti batch viewer show batch singl video stream share multipl concurr viewer resum request wait addit stream capac actual resumpt occur convent approach support demand paus resum video access stream disk video request greatli increas disk arm requir vod system paper propos effici mechan support paus resum featur ahead schedul buffer idea buffer improv numb character databas access pattern analyt predict buffer hit probabl abstract analyt predict buffer hit probabl base charac teriz databas access real refer trace extrem workload manag system capac plan knowledg ful proper alloc buffer space databas relat manag buffer space mix transact queri environ access character predict buffer invalid effect multi node environ turn influenc transact rout strate gi challeng character databas access pattern real workload refer trace simpl manner easili comput buffer hit probabl articl character method distin guish type access pattern trace local transact random access transact sequenti access long queri propos concis character access skew randomli access page logic group larg number data page small number partit frequenc access page par tition treat equal base approach present recurs binari partit algorithm infer access skew character buffer hit probabl subset buffer size valid buffer hit predict singl multipl node system product databas trace show propos approach predict buffer hit probabl composit workload compon file kei word databas access character access skew sequenti access ref erenc trace workload manag analyt predict support fulli interact playout disk darrai dbase video server multivers diverg control time fuzzi epsilon serializ esr propos manag control inconsist extend classic transact process base serializ esr increas transact process system concurr toler bound amount inconsist paper present multivers diverg control mvdc algorithm support esr time fuzzi multivers databas unlik fuzzi accumul time fuzzi semant simpl summat length time interv underestim total time fuzzi result incorrect execut overestim total time fuzzi unnecessarili degrad effect mvesr paper present oper call timeunion accur accumul total time fuzzi addit describ mvdc algorithm correctli bound inconsist time mvesr ac data placement buffer manag concurr mergesort parallel prefetch parallel transact process coupl system paper conduct perform studi parallel transact process coupl system coupl system multi node system share global buffer develop multiplesystem simul obtain perform result simul run workload coupl system behavior input studi statist includ local global buffer hit page write global buffer cross invalid castout compar analyz relationship degre data skew explor introduct due rapid technolog advanc recent attract coupl multipl system databas transact process order address increas demand capac avail cost perform method coupl multipl system data share approach refer dynam pariti group improv write perform raid disk arrai schedul multipl queri parallel machin good deal progress made recent effici parallel individu phase singl queri multiprocessor databas system paper devis evalu number schedul algorithm design handl multipl parallel queri algorithm emerg clear winner algorithm hierarch natur phase good qualiti preced base schedul creat individu queri number processor compon emploi dynam program phase result phase creat schedul full set queri compon base previous publish work nonpreced base malleabl schedul problem np hard strong sens multipl queri schedul gener hierarch algorithm experiment achiev result close optim introdu parallel execut multipl pipelin hash join paper studi parallel execut multipl pipelin hash join specif deal issu processor alloc hash filter improv parallel execut hash join present scheme transform bushi execut tree alloc tree node denot pipelin processor alloc node alloc tree base concept synchron execut time relat hash tabl pipelin made approxim time addit approach hash filter investig improv perform perform studi conduct simul demonstr import processor alloc evalu scheme hash filter simul result processor alloc base alloc tree significantli outperform base origin bushi tree effect hash filt asymtot perform buffer model data share environ parallel execut multipl pipelin hash join paper studi parallel execut multipl pipelin hash join specif deal issu processor alloc hash filter improv parallel execut hash join present scheme transform bushi execut tree alloc tree node denot pipelin processor alloc node alloc tree base concept synchron execut time relat hash tabl pipelin made approxim time addit approach hash filter investig improv perform perform studi conduct simul demonstr import processor alloc evalu scheme hash filter simul result processor alloc base alloc tree significantli outperform base origin bushi tree effect hash filt schedul parallel task minim averag respons time interv order partial order defin interv represent transit orient digraph interv order iff undirect complement chordal investig parallel algorithm schedul problem system consist set task requir unit execut time interv order ident parallel processor construct optim minim length schedul algorithm base subroutin comput call schedul distanc minim number time step need schedul task succeed task preced task interv order task schedul distanc comput processor log time crew pram give increment version schedul distanc algorithm comput empti slot schedul paralleliz task minim averag respons time index select scheme nest object hierarchi paper address problem devis set index nest object hierarchi object orient databas improv system perform note effect index entangl inclus index affect benefit achiev index phenomenon term index interact effect index interact consider set index built index select problem formul index select algorithm evalu simul effect object function guid search index select algorithm investig shown simul result greedi algorithm devis light phenomenon index interact perform fairli case sensit analysi databas paramet conduct index term object orient databas index nest object hierarchi index interact perform model comparison global share buffer manag polici cluster environ graph theoret approach determin join reduc sequenc distribut queri process buffer analysi data share environ skew data access perform analysi affin cluster transact process coupl architectur algorithm parallel relat databas join presenc data skew optim nodup dto dall broadcast scheme distribut comput system perform studi robust distribut load share strategi perform evalu transact process coupl architectur handl system dynam analyt model reconstruct time mirror disk optim group sweep schedul gss heterogen multimedia stream parallel queri process relat grow larger queri complex parallel queri process increasingli attract option improv perform databas system object paper examin issu encount parallel queri process techniqu address issu focu paper join oper sort merg join hash join consid type parallel exploit intra oper inter oper inter queri parallel intra oper parallel major issu task creation object split join oper task manner load spread evenli number processor challeng valu join attribut uniformli distribut inter oper parallel achiev parallel execut independ oper pipelin case decentr consensu protocol multi dport commun develop paper effici decentr consensu protocol distribut system multiport commun arbitrari number node system propos protocol reach consensu minim number messag step specif develop class decentr consensu protocol initi initi present case port commun node send messag step show algorithm develop result case multi port commun port commun mean node send messag step deriv algorithm gener version algorithm present paper optim sens complet protocol minim number messag step requir number messag incur algorithm deriv distribut diverg control epsilon serializ databas access character buffer hit predict dynam finit version effect version approach concurr transact queri process sever data content exist short updat transact long read queri access databas convent concurr control method paper present dynam finit version dfv scheme effect support concurr process transact queri acquir lock queri read small fix number dynam deriv transact consist slightli obsolet logic snapshot databas hand transact access date data databas data content queri intermedi version creat snapshot automat discard dirti page updat activ transact allow written back databas commit steal polici time consist logic snapshot advanc automat quiesc ongo transact queri introduct perform analysi dual stripe strategi replic disk arrai issu design multi dserver file system cope load skew model analysi transact process system standard lock phase lock demand lock request block lock conflict primari concurr control cc method central databas main sourc perform degrad standard lock block transact txn restart resolv deadlock secondari effect perform provid perform analysi standard lock accur estim perform degrad lead thrash introduc set method cope perform limit restartori lock method select abort txn increas level concurr activ txn respect standard lock high content environ run prioriti method abort block txn base essenti block principl block activ txn waitdepth limit wdl method minim wast process base abort decis progress made txn restart wait serv load control mechan defer restart abort txn conflict txn left system method perform superior restartori perform comparison thrash control polici concurr mergesort parallel prefetch optim processor alloc support pipelin hash join paper develop algorithm achiev optim processor alloc pipelin hash join multiprocessorbas databas system pipelin hash join compos stage join oper pipelin execut phase tabl build phase tupl probe phase focu problem alloc processor stage pipelin minim queri execut time formul processor alloc problem twophas mini max optim problem develop optim alloc scheme constraint effect problem formul solut verifi detail tupl tupl simul pipelin hash join solut scheme gener applic optim resourc alloc problem formul phase mini max problem introduct recent year multiprocessor base parallel databas machin appli hash filter improv execut bushi tree paper explor approach interleav bushi execut tree hash filter improv execut multi join queri effect hash filter evalu effici scheme determin effect sequenc hash filter bushi execut tree develop hash filter built appli base join sequenc bushi tree reduct effect optim cost minim scheme hash filter implement evalu simul experiment shown applic hash filter gener power mean improv execut multi join queri improv promin number relat queri increas introduct relat databas system join expens oper execut applic involv decis support complex object speci perform comparison dynam polici remot cach combin join semi djoin oper distribut queri process parallel sort merg join algorithm manag data skew replic algorithm remot cach architectur parallel hash join algorithm manag data skew perform analysi buffer coher polici multisystem data share environ perform analysi concurr control lock defer block analyt model databas concurr control group sweep schedul dasd dbase multimedia storag manag buffer manag base return consumpt multi dqueri environ design model cluster raid perform comparison activ dsender activ dreceiv polici distribut cach effici decentr consensu protocol distribut comput system comparison regress dbase load share strategi distribut databas environ effect system dynam coupl architectur transact process diverg control epsilon dserializ distribut object replic strategi remot cach architectur schedul processor alloc parallel execut multi djoin queri distribut object replic strategi remot cach architectur mon dsong chen futur advanc network storag enabl wide spectrum comput mediat structur collabor individu paper present model captur divers type structur collabor model combin effici power hierarchi abstract lowest level stream media commun modul access right particip collabor higher level abstract session repres collect semant relat media stream confer repres tempor relat sequenc session abstract model support unif synchron asynchron collabor sophist access control intra group inter group collabor yield power set build block construct multimedia applic rich environ carri structur multimedia collabor keyword perform analysi dynam finit version concurr transact queri process paper analyz perform dynam finit version dfv scheme concurr transact queri process finit number consist snapshot deriv queri access develop analyt model base renew process approxim evalu perform dfv snapshot storag overhead obsolesc face queri measur simul valid analyt model evalu trade off strategi advanc snapshot result show analyt model match close simul storag overhead obsolesc sensit snapshot advanc strategi snapshot gener speak increas number snapshot demonstr trade storag overhead queri obsolesc case skew access low updat rate moder increas number snapshot schedul paralleliz task put shelf character databas access skew transact process environ perform analysi coher control polici lock retent approxim algorithm schedul paralleliz task impact workload partition perform coupl architectur transact process segment ddeep tree execut pipelin hash join paper explor execut pipelin hash join multiprocessor base databas system improv queri execut innov ap proach queri execut tree select propos exploit segment deep tree bushi tree deep subtre deriv analyt ical model execut pipelin segment light model develop heurist scheme determin queri execut plan base seg ment deep tree queri ef ficient execut shown simul propos approach incur addit head plan execut possess flexibl queri plan gener lead queri plan sig nificantli perform achiev previou scheme deep tree interleav join sequenc semijoin distribut queri process distribut queri process convent approach reduc amount data transmiss appli sequenc semijoin reduc ship result relat final site carri join oper recent shown approach appli combin join semijoin reduc lead substanti larger reduct data transmiss requir paper develop effici heurist approach determin effect sequenc semijoin join reduc semijoin execut reduc amount data transmiss requir perform join sequenc term benefici semijoin join sequenc note benefici semijoin includ convent profit semijoin gain semijoin profit benefici due inclus join reduc type depend semijoin join reduc complic identificati analysi concurr dcoher control protocol distribut transact process system region local analysi hybrid concurr control scheme high data content environ workload character relat databas environ relat databas system increasingli popular clear understand workload system design tune effect relat databas workload analyz redwar develop character workload db environ appli studi product db system sql trace hour interv imag copi databas catalog obtain result workload studi summar focu structur complex sql statement makeup run time behavior transact queri composit relat view result obtain provid import inform need build benchmark workload evalu altern design trade off databas system index term relat databas workload character db introduct relat databas system grow increasingli popular throughput respons time requir becomin perform comparison buffer coher polici effect algorithm parallel hash join presenc data skew determin benefici semijoin join sequenc distribut queri process optim buffer alloc multi dqueri environ dynam load share presenc inform obsolesc distribut databas environ polici effici resourc util remot cach architectur optim dto dall broadcast scheme distribut system compar perform parallel join algorithm analyt model hierarch buffer data share environ model analysi time dstamp histori base certif protocol concurr control adapt strategi load share distribut databas environ inform lag effect algorithm parallel sort merg presenc data skew join oper reduc distribut queri process combin join semijoin oper distribut queri process concurr control lock defer block databas buffer model data share environ workload character relat databas environ relat databas system increasingli popular clear understand workload system design tune effect relat databas workload analyz redwar develop character workload db environ appli studi product db system sql trace hour interv imag copi databas catalog obtain result workload studi summar focu structur complex sql statement makeup run time behavior transact queri composit relat view result obtain provid import inform need build benchmark workload evalu altern design trade off databas system index term relat databas workload character db introduct relat databas system grow increasingli popular throughput respons time requir becomin effect skew data access buffer hit data content data share environ effect parallel process databas system integr approach buffer manag queri optim analysi replic distribut databas system effect parallel join join expens frequent oper parallel highli desir effect parallel join depend abil evenli divid load processor data skew disastr effect perform skew handl algorithm propos remain gener ineffici case multi join due join product skew costli unnecessari redistribut commun cost parallel join algorithm call fa join introduc earlier paper determinist perfect balanc properti advantag fa join sensit correl attribut distribut relat present improv version algorithm call sfa join symmetr treatment relat predict low join product attribut skew make suitabl repeat multi join oper perform analyz theoret exp effect approach vertic partit physic design relat databas adapt transact rout heterogen databas environ limit factor join perform parallel processor analysi parallel process architectur databas system perform analysi optimist concurr control scheme system larg memori perform comparis concurr control protocol transact process system region local integr buffer manag queri optim relat databas environ robust transact drout strategi distribut databas system integr concurr dcoher control multisystem data share multisystem coupl combin data share data partit multisystem coupl combin data share data partit optim site assign relat distribut databas environ perform comparison io ship databas call ship scheme multisystem partit databas effect skew join perform parallel architectur robust transact rout distribut databas system load share hybrid distribut central databas system site assign relat join oper distribut transact process environ hybrid data share data partit architectur transact process perform comparison configur duplex system tradeoff coupl small larg processor transact process optim design sequenti analysi vlsi test strategi adapt optim model applic test vlsi circuit test adapt optim model relat assign distribut transact process environ central versu geograph distribut databas system vertic partit algorithm relat databas design analysi integr concurr dcoher control analysi fault toler multiprocessor architectur lock engin design progress transact recoveri distribut db fdc system analysi affin base rout multi dsystem data share dynam load share distribut databas system perform analysi oper mode configur duplex system coupl partit databas system analysi multi dsystem function request ship analysi recoveri protocol distribut dline transact process system coupl small system transact process coupl small system transact process transact recoveri distribut db fdc system progress approach affin base rout multi dsystem data share multisystem coupl function request ship distribut concurr control analysi data share model central concurr control multi dsystem environ 