Robo-line Storage: Low Latency, High Capacity Storage Systems over Geographically Distributed Networks

Rapid advances inhighperformancecomputing aremaking possiblemore complete and accuratecomputer-basedmodeling of complex physicalphenomena, such as weather frontinteractions, dynamics ofchemicalreactions, numericalaerodynamicanalysisofairframes, and ocean-land-atmosphere interactions. Many of these"grand challenge"applications areas demanding of the underlyingstorage system,intermsoftheircapacityand bandwidthrequirements, astheyareon thecomputationalpower of theprocessor. A globalview oftheEarth'soceanchlorophylland landvegetationrequiresover2 terabytes of raw satellite image data[ISTP9I]! Inthispaper,we describeourplannedresearchprogram inhighcapacity, highbandwidthstoragesystems.The projecthas fouroverallgoals.First, we willexamine new methods forhighcapacitystoragesystems,made possibleby low cost,smallformfactormagneticand opticaltapesystems.Second,accesstothe storagesystemwillbe low latencyand highbandwidth.To achievethis,we must inteflcave datatransfer at alllevelsofthestoragesystem,includingdevices,controllers, ervers, and communicationslinks.Latency willbe reducedby extensivecachingthroughoutthestoragehierarchy. Third,we willprovideeffective management of a storagehierarchy,extendingthe techniquesalreadydevelopedby OusterhoutforhisLog Structured FileSystem.Finally, we willconstructaprotototype highcapacityfileserver,suitable foruseon theNationalResearchand EducationNetwork (NREN). Such researchmust be acornerstoneofany coherentprogram inhighperformancecomputingand communications.