CS 267 Applications of Parallel ComputersSupercomputing:The Past and FutureKathy Yelickwww.cs.berkeley.edu/~yelick/cs267_s07Outline• Historical perspective (1985 to 2005)from Horst Simon• Recent past: what’s new in 2007• Major challenges and opportunities forthe futureSlide source: Horst SimonSignpost System 1985Cray-2• 244 MHz (4.1 nsec)• 4 processors• 1.95 Gflop/s peak• 2 GB memory (256 MW)• 1.2 Gflop/s LINPACK R_max• 1.6 m2 floor space• 0.2 MW powerSlide source: Horst SimonSignpost System in 2005IBM BG/L @ LLNL• 700 MHz (x 2.86)• 65,536 nodes (x 16,384)• 180 (360) Tflop/s peak (x 92,307)• 32 TB memory (x 16,000)• 135 Tflop/s LINPACK (x 110,000)• 250 m2 floor space (x 156)• 1.8 MW power (x 9)Slide source: Horst Simon1985 versus 2005• custom built vector mainframeplatforms• 30 Mflops sustained is goodperformance• vector Fortran• proprietary operating system• remote batch only• no visualization• no tools, hand tuning only• dumb terminals• remote access via 9600 baud• single software developer,develops and codes everything• serial, vectorized algorithms• commodity massively parallelplatforms• 1 Tflops sustained is goodperformance• Fortan/C with MPI, object orientation• Unix, Linux• interactive use• visualization• parallel debugger, development tools• high performance desktop• remote access via 10 Gb/s; grid tools• large group developed software,code share and reuse• parallel algorithmsSlide source: Horst SimonThe Top 10 Major Accomplishmentsin Supercomputing in the Past 20 Years•Horst’s Simon’s list from 2005• Selected by “impact” and “change in perspective”10) The TOP500 list9) NAS Parallel Benchmarks8) The “grid”7) Hierarchical algorithms: multigrid and fast multipole6) HPCC initiative and Grand Challenge application 5) Attack of the killer microsSlide source: Horst Simon- Listing of the 500 most powerful Computers in the World - Yardstick: Rmax from LinpackAx=b, dense problem - Updated twice a year:ISC‘xy in Germany, June xySC‘xy in USA, November xy - All data available from www.top500.org - Good and bad effects of this list/competitionSizeRateTPP performance#10) TOP500TOP500 list - Data shownTOP500 list - Data shown• Manufacturer Manufacturer or vendor• Computer Type indicated by manufacturer or vendor• Installation Site Customer• Location Location and country• Year Year of installation/last major update• Customer Segment Academic,Research,Industry,Vendor,Class.• # Processors Number of processors• Rmax Maxmimal LINPACK performance achieved• RpeakTheoretical peak performance• NmaxProblemsize for achieving Rmax• N1/2Problemsize for achieving half of Rmax• Nworld Position within the TOP500 ranking1001000100001000001E+061E+071E+081E+091E+101E+111E+121993 1996 1999 2002 2005 2008 2011 2014SUM#1#500Petaflop with ~1M Cores By 20081Eflop/s100 Pflop/s10 Pflop/s1 Pflop/s100 Tflop/s10 Tflops/s1 Tflop/s100 Gflop/s10 Gflop/s1 Gflop/s10 MFlop/s1 PFlop system in 2008Data from top500.org6-8 yearsCommon by 2015?Slide source: Horst SimonPetaflop with ~1M Cores in your PC by 2025?#4 Beowulf Clusters• Thomas Sterling et al. established visionof low cost, high end computing• Demonstrated effectiveness of PCclusters for some (not all) classesof applications• Provided software and conveyed findingsto broad community (great PR) throughtutorials and book (1999)• Made parallel computing accessible tolarge community worldwide; broadenedand democratized HPC; increaseddemand for HPC• However, effectively stopped HPCarchitecture innovation for at least adecade; narrower market for commoditysystemsSlide source: Horst Simon#3 Scientific Visualization• NSF Report, “Visualization inScientific Computing” establishedthe field in 1987 (edited by B.H. McCormick,T.A. DeFanti, and M.D. Brown)• Change in point of view:transformed computer graphicsfrom a technology driven subfieldof computer science into a mediumfor communication• Added artistic element• The role of visualization is “toreveal concepts that are otherwiseinvisible” (Krystof Lenk)Slide source: Horst SimonBefore Scientific Visualization (1985)Computer graphicstypical of the time:– 2 dimensional– line drawings– black and white– “vectors” used todisplay vector fieldImages from a CFD report atBoeing (1985).Slide source: Horst SimonAfter scientific visualization (1992)The impact of scientific visualization seven years later:– 3 dimensional– use of “ribbons” and “tracers” to visualize flow field– color used to characterize updraft and downdraftImages from “Supercomputing and the Transformation of Science” byKauffmanand Smarr, 1992; visualization by NCSA; simulation by BobWilhelmson, NCSASlide source: Horst Simon#2 Message Passing Interface (MPI)MPISlide source: Horst SimonParallel Programming 1988• At the 1988 “Salishan” conferencethere was a bake-off of parallelprogramming languages trying tosolve five scientific problems• The “Salishan Problems” (ed. JohnFeo, published 1992) investigatedfour programming languages– Sisal, Haskel, Unity, LGDF• Significant research activity at thetime• The early work on parallellanguages is all but forgottentodaySlide source: Horst SimonParallel Programming 1990• The availability of real parallelmachines moved thediscussion from the domainof theoretical CS to thepragmatical application area• In this presentation (ca. 1990)Jack Dongarra lists sixapproaches to parallelprocessing• Note that message passinglibraries are a sub-item on 2)Slide source: Horst SimonParallel Programming 1994#1 Scaled Speed-UpThe argument against massiveparallelism (ca. 1988)Slide source: Horst SimonAmdahl’s Law: speed = base_speed / ( (1-f) + f/nprocs ) infinitely parallel Cray YMP base_speed = .1 2.4 nprocs = infinity 8Then speed(Infinitely Parallel) > speed(Cray) only if f > .994Challenges for the Future• Petascale computing• Multicore and the memory wall• Performance understanding at scale• Topology-sensitive interconnects• Programming models for the massesApplication Status in 2005• A few Teraflop/s sustained performance• Scaled to 512 - 1024 processorsParallel jobsize atNERSCHow to Waste Machine $8-byte Roundtrip
View Full Document
Unlocking...