Memory Consistency Models in Wide-area Storage System – Or What Do They Mean?MotivationsData and Computation Hungry ApplicationsData Grid for High Energy Physics - CalTechBackgroundRunning Parallel Applications on OceanStorePowerPoint PresentationRunning SMP Apps on OceanStoreConsistency ModelsPerformance EvaluationComputation TimeNetwork LatencyOpen QuestionsConclusion and Future WorkMemory Consistency Models in Wide-area Storage System – Or What Do They Mean?CS258 Spring 2002Mark Whitney and Yitao DuanMotivations•Global scale computing approaching•Wide area storage is becoming a reality•The greed for processing power calls the marriage of the two•Traditional approach to large scale data processing: Hierarchy•What if new algorithms require to touch more data? Scale SMP?•Use OceanStore as testbedData and Computation Hungry Applications•Quantum Chromodynamics•Biomolecular Dynamics•Weather Forecasting•Cosmological Dark Matter•Biomolecular Electrostatics•Electric and Magnetic Molecular PropertiesData Grid for High Energy Physics - CalTechTier2 Centre ~1 TIPSOnline SystemOffline Processor Farm ~20 TIPSCERN Computer CentreFermiLab ~4 TIPSFrance Regional Centre Italy Regional Centre Germany Regional Centre InstituteInstituteInstituteInstitute ~0.25TIPSPhysicist workstations~100 MBytes/sec~100 MBytes/sec~622 Mbits/sec~1 MBytes/secThere is a “bunch crossing” every 25 nsecs.There are 100 “triggers” per secondEach triggered event is ~1 MByte in sizePhysicists work on analysis “channels”.Each institute will have ~10 physicists working on one or more channels; data for these channels should be cached by the institute serverPhysics data cache~PBytes/sec ~622 Mbits/sec or Air Freight (deprecated)Tier2 Centre ~1 TIPSTier2 Centre ~1 TIPSTier2 Centre ~1 TIPSCaltech ~1 TIPS~622 Mbits/secTier 0Tier 0Tier 1Tier 1Tier 2Tier 2Tier 4Tier 41 TIPS is approximately 25,000 SpecInt95 equivalentsBackground•What is OceanStore?–A global persistent data store scalable to billions of users–High availability, fault-tolerance, security–Caching to reduce network congestion, guarantee availability and performance–Flexible consistency semantics•Observations and questions–Remarkable resemblance to MP memory system•Replica = cache, client = processor, data object = memory item–OceanStore consistency semantics are typically that of a file system’s. What do they mean to a program?Running Parallel Applications on OceanStore•Why do we try this–Distributed computing–Grid–World Wide Computing–New programming paradigm? (OceanStore is a new phenomena, will it bring out new applications? Where will computing infrastructure go given the advance of network, storage and parallel processing?ParaAppOS KernelOClientParaAppOS KernelOClientParaAppOS KernelOClientShared Virtual Memory SpaceRunning SMP Apps on OceanStore•OceanStore data objects are globally identified•Virtual address in application address space mapped to OceanStore object ID•Shared memory address access turned into OceanStore requestsConsistency ModelsPerformance Evaluation•OceanStore …(#of inner rings, …)•Nachos++!•MIPS R3000 processor w/FP•Stanford SPLASH-2 benchmark suite•4 x 4 matrix LU decompositionComputation Time6400066000680007000072000740007600078000800008200084000Blocksize = 2 Blocksize = 41 Processor4 Processors (SC)4 Processors (RC)Number of CyclesNetwork Latency020000040000060000080000010000001200000Blocksize = 2 Blocksize = 41 Processor4 Processors(SC)4 processors(RC)MillisecondsOpen Questions•Programming model•Cache policy•Consistency models•Sharing granularityConclusion and Future Work•Matrix decomposition runs on OceanStore!•Wide-area distributed synchronizations are expensive (not surprising)•Need better memory model if want to run shared memory applications•Message passing? – Seems to be a better match (use explicit OceanStore APIs)•New programming
View Full Document
Unlocking...