Grid ComputingOutlineBack to the FutureProcessor EvolutionSlide 5Slide 6From single-CPU to Multi-CoreSlide 8Multi-Core – The Pit FallsCostsSlide 11Lessons from History - SummaryParallelism through Software - GridGrid InfrastructureSlide 15Slide 16Grid Components - HardwareGrid Components – SoftwareGrid Components StackGrid WorkflowSlide 21Grid – Resource ManagementGrid Resource Management - GlobusSlide 24Slide 25Slide 26Grid SchedulingGrid SchedulingSlide 29Slide 30Grid Scheduling - CondorGrid Scheduling – CondorSlide 33Slide 34Slide 35Grid Scheduling – CondorSlide 37Open IssuesSlide 39Slide 40ApplicationsAcknowledgementGrid ComputingGrid ComputingAdvance Computer Advance Computer ArchitectureArchitectureCSE 8383CSE 8383OutlineOutlineThe Motivation Behind GridThe Motivation Behind GridExisting Grid InfrastructuresExisting Grid InfrastructuresIntegral Components of a Grid Integral Components of a Grid System System Open IssuesOpen IssuesBack to the FutureBack to the Future““The complexity for minimum The complexity for minimum component costs has increased component costs has increased at a rate of roughly a factor of at a rate of roughly a factor of two per year…”two per year…”… … circuit densities in chips would circuit densities in chips would double every 12-18 months.double every 12-18 months.Gordon Moore, Electronics Magazine. April Gordon Moore, Electronics Magazine. April 19, 196519, 1965Processor EvolutionProcessor EvolutionFrom Processor Generation n to n+1,From Processor Generation n to n+1,Gate delay reduces by 1/Gate delay reduces by 1/2 (basic frequency 2 (basic frequency goes up by goes up by 2)2)Number of transistors in a constant area (cost) Number of transistors in a constant area (cost) goes up by 2goes up by 2Additional transistors enable an additional Additional transistors enable an additional increase in performance by the factor of increase in performance by the factor of 22Deeper pipeline, Offset pipeline penalties (branch Deeper pipeline, Offset pipeline penalties (branch prediction), Increased number and/or size of caches, prediction), Increased number and/or size of caches, Exploit Instruction Level Parallelism (ILP)Exploit Instruction Level Parallelism (ILP)Result: 2x Performance at roughly the Result: 2x Performance at roughly the equal costequal costProcessor EvolutionProcessor EvolutionOver the past 37 Over the past 37 years, Moore’s law years, Moore’s law has successfully has successfully predicted the predicted the exponential growth exponential growth of component of component densitiesdensitiesYearYearTransistoTransistorsrs197019702,3002,300197519756,0006,0001980198029,00029,00019851985275,000275,000199019905,500,0005,500,0002000200042,000,0042,000,0000Processor EvolutionProcessor EvolutionMoore’s law heldMoore’s law heldMainly because of advances in micro-Mainly because of advances in micro-architecture that exploited huge growth in architecture that exploited huge growth in transistor/area and that overcame interconnect transistor/area and that overcame interconnect limitationslimitationsScarcity of new micro-architecture ideasScarcity of new micro-architecture ideasPipelining, branch prediction, IP, Caching, has Pipelining, branch prediction, IP, Caching, has reached a point of maturityreached a point of maturityPower / Current IssuesPower / Current IssuesPower density limitationsPower density limitationsVoltage and doping (less switching energy Voltage and doping (less switching energy avaialable) avaialable)From single-CPU to From single-CPU to Multi-CoreMulti-CoreGate delay does not reduce much (basic Gate delay does not reduce much (basic frequency goes up a little)frequency goes up a little)Number of transistors in a constant area Number of transistors in a constant area (cost) goes up by 2(cost) goes up by 2Global Wiring is not stressedGlobal Wiring is not stressedResult: 2x performance at roughly equal costResult: 2x performance at roughly equal costFrom single-CPU to From single-CPU to Multi-CoreMulti-CoreOn the order of 1000, “classic” On the order of 1000, “classic” pipelined CPUs can be fit in the area pipelined CPUs can be fit in the area ofof a “state-of-the-art” high-end CPU.a “state-of-the-art” high-end CPU.Can have the same performance as Can have the same performance as high end CPUs but with limited scopehigh end CPUs but with limited scopeResult: ~1000x performance at Result: ~1000x performance at roughly equal the costroughly equal the costIf dual-core is 2x, quad-core is 4x, then If dual-core is 2x, quad-core is 4x, then why not?why not?Multi-Core – The Pit FallsMulti-Core – The Pit FallsAmdahl’s LawAmdahl’s LawA Portion of all parallel execution is serializedA Portion of all parallel execution is serializedCommunication – dependence on uncompleted parallel Communication – dependence on uncompleted parallel executionexecutionConflicts – Shared ResourcesConflicts – Shared ResourcesCoordination – Synchronization with other executionsCoordination – Synchronization with other executionsOverhead – For above mechanisms (semaphores, locks, Overhead – For above mechanisms (semaphores, locks, etc)etc)Speedup = Speedup = 1/(%serial + (1-%serial)/#CPUs)1/(%serial + (1-%serial)/#CPUs)All parallel computations are uniform and All parallel computations are uniform and take equal timetake equal timeCosts Costs ““The cost of manufacturing facilities doubles every The cost of manufacturing facilities doubles every generation. In the late 1980s, billion-dollar plants seemed generation. In the late 1980s, billion-dollar plants seemed like something a long way in the future. They seemed like something a long way in the future. They seemed almost inconceivable. But now, Intel has two plants that almost inconceivable. But now, Intel has two plants that will cost more than will cost more than $2.5 billion$2.5 billion. If we double it . If we double it for a for a couple of generationscouple of generations, we’re looking at , we’re looking at $10 billion $10 billion plantsplants. I don’t think there’s any industry in the world that . I don’t think there’s any industry in the world that builds $10 billion plants, although oil refineries probably builds $10 billion plants, although oil refineries probably come closecome close “