mine misconfigur machin grid system noam palatin ari leizarowitz assaf schuster ran wolff grid system prove increasingli manag batch comput job organ intel intern develop netbatch system manag ten thousand machin size heterogen complex grid system make difficult configur result misconfigur machin advers affect entir system investig distribut data mine approach detect misconfigur machin grid monitor system gm intrus collect data sourc log file system servic grid system convert raw data semant meaning data store data machin obtain limit incur overhead allow scalabl analysi request distribut outlier detect algorithm emploi identifi misconfigur machin algorithm implement recurs workflow grid job suit grid system machin unavail time fail altogeth 