High Performance Computing, Computational Grid, and ...

1

1

High Performance Computing, High Performance Computing, Computational Grid, and Numerical LibrariesComputational Grid, and Numerical Libraries

Jack DongarraInnovative Computing LabUniversity of Tennessee

http://http://www.cs.utk.edu/~dongarrawww.cs.utk.edu/~dongarra//

The 14th Symposium on Computer Architecture and High Performance Computing

Vitoria/ES - Brazil - October 28-30, 2002

2

Technology Trends: Technology Trends: Microprocessor CapacityMicroprocessor Capacity

2X transistors/Chip Every 1.5 yearsCalled “Moore’s Law”

Moore’s Law

Microprocessors have become smaller, denser, and more powerful.Not just processors, bandwidth, storage, etc

Gordon Moore (co-founder of Intel) predicted in 1965 that the transistor density of semiconductor chips would double roughly every 18 months.

2

3

2005

2010

ASCI WhitePacific

EDSAC 1UNIVAC 1

IBM 7090

CDC 6600

IBM 360/195CDC 7600

Cray 1

Cray X-MPCray 2

TMC CM-2

TMC CM-5 Cray T3D

ASCI Red

1950 1960 1970 1980 1990 2000 2010

1 KFlop/s

1 MFlop/s

1 GFlop/s

1 TFlop/s

1 PFlop/s

Scalar

Super Scalar

Vector

Parallel

Super Scalar/Vector/ParallelMoore’s LawMoore’s Law

4

H. Meuer, H. Simon, E. Strohmaier, & JDH. Meuer, H. Simon, E. Strohmaier, & JD

- Listing of the 500 most powerfulComputers in the World

- Yardstick: Rmax from LINPACK MPPAx=b, dense problem

- Updated twice a yearSC‘xy in the States in NovemberMeeting in Mannheim, Germany in June

- All data available from www.top500.org

Size

Rat

e

TPP performance

3

5

Fastest Computer Over Time

0

10

20

30

40

50

60

70

1990 1992 1994 1996 1998 2000

Year

GF

lop/

s

X Y ( S c a t t e r ) 1

Cray Y-MP (8)

TMCCM-2(2048)

Fujitsu VP-2600

In 1980 a computation that took 1 full year to complete can now be done in ~ 10 hours!

6


TMCCM-2(2048)

Fujitsu VP-2600

Cray Y-MP (8)

0

100

200

300

400

500

600

700

1990 1992 1994 1996 1998 2000

Year

GF

lop/

s

X Y ( S c a t t e r ) 1

HitachiCP-PACS(2040)

IntelParagon(6788)

FujitsuVPP-500(140)

TMC CM-5(1024)NEC

SX-3(4)

In 1980 a computation that took 1 full year to completecan now be done in ~ 16 minutes!

4

7


HitachiCP-PACS

(2040)

IntelParagon(6788)

FujitsuVPP-500

(140)

TMC CM-5(1024)

NEC SX-3(4)

TMCCM-2(2048)

Fujitsu VP-2600

Cray Y-MP (8)

0

1000

2000

3000

4000

5000

6000

7000

1990 1992 1994 1996 1998 2000

Year

GF

lop/

s

X Y ( S c a t t e r ) 1

ASCI White Pacific(7424)

Intel ASCI Red Xeon(9632)

SGI ASCIBlue Mountain(5040)

Intel ASCI Red(9152)

ASCI BluePacific SST(5808)

In 1980 a computation that took 1 full year to completecan today be done in ~ 27 seconds!

8


HitachiCP-PACS

(2040)

IntelParagon(6788)

FujitsuVPP-500

(140)

TMC CM-5(1024)

NEC SX-3(4)

TMCCM-2(2048)

Fujitsu VP-2600

Cray Y-MP (8)

0

10

20

30

40

50

60

70

1990 1992 1994 1996 1998 2000

Year

TF

lop/

s

XY (Scatter) 1

2002

Intel ASCIRed(9152)

ASCI WhitePacific(7424)

Intel ASCIRed Xeon

(9632)ASCI BlueMountain

(5040)

JapaneseEarth SimulatorNEC 5104

In 1980 a computation that took 1 full year to completecan today be done in ~ 5.4 seconds!

5

9

Machines at the Top of the List Machines at the Top of the List

53%140236124.5Fujitsu NWT1993

83%67681.43382.3281.1Intel Paragon XP/S MP1994

83%67681.03381281.1Intel Paragon XP/S MP1995

60%20481.86141.3368.2Hitachi CP-PACS1996

73%91523.018303.61338Intel ASCI Option Red (200 MHz Pentium Pro)

1997

55%58082.138681.62144ASCI Blue-Pacific SST, IBM SP 604E

1998

74%96320.832071.12379ASCI Red Intel Pentium II Xeon core

1999

44%74243.5111362.14938ASCI White -Pacific, IBM SP Power 3

2000

65%74241.0111361.57226ASCI White -Pacific, IBM SP Power 3

2001

88%51203.7409605.035860Earth Simulator Computer, NEC

2002

EfficiencyNumber ofProcessors

Factor ?from Pervious Year

TheoreticalPeak Gflop/s

Factor ?from Pervious Year

MeasuredGflop/s

ComputerYear

10

A Tour A Tour dd’’ForceForce in Engineeringin Engineering♦ Homogeneous, Centralized,

Proprietary, Expensive! ♦ Target Application: CFD-Weather,

Climate, Earthquakes♦ 640 NEC SX/6 Nodes (mod)

Ø 5120 CPUs which have vector ops♦ 40TeraFlops (peak)♦ $250-$500 million for things in

building♦ Footprint of 4 tennis courts♦ 7 MWatts

Ø Say 10 cent/KWhr - $16.8K/day = $6M/year!

♦ Expect to be on top of Top500 until 60-100 TFlop ASCI machine arrives

♦ For the Top500 (June 2002)Ø Equivalent ~ 1/6 S Top 500 Ø Performance of ESC

> S Next Top 12 ComputersØ S of all the DOE computers

= 27.5 TFlop/sØ Performance of ESC

> All the DOE + DOD machines (37.2 TFlop/s)

6

11

Top10 of the Top500Top10 of the Top500Rank Manufacturer Computer

Rmax

[TF/s] Installation Site Country Year Area of

Installation # Proc

1 NEC Earth-Simulator 35.86 Earth Simulator Center Japan 2002 Research 5120

2 IBM ASCI White

SP Power3 7.23 Lawrence Livermore

National Laboratory USA 2000 Research 8192

3 HP AlphaServer SC ES45 1 GHz 4.46 Pittsburgh

Supercomputing Center USA 2001 Academic 3016

4 HP

AlphaServer SC ES45 1 GHz 3.98

Commissariat a l’Energie Atomique (CEA) France 2001 Research 2560

5 IBM SP Power3 375 MHz 3.05 NERSC/LBNL USA 2001 Research 3328

6 HP AlphaServer SC ES45 1 GHz

2.92 Los Alamos National Laboratory

USA 2002 Research 2048

7 Intel ASCI Red 2.38 Sandia National Laboratory USA 1999 Research 9632

8 IBM pSeries 690 1.3 GHz

2.31 Oak Ridge National Laboratory


9 IBM ASCI Blue Pacific SST, IBM SP 604e

2.14 Lawrence Livermore National Laboratory


10 IBM pSeries 690 1.3 Ghz 2.00 IBM/US Army Reseach Lab

(ARL) USA 2002 Vendor 768

12

TOP500 TOP500 -- PerformancePerformance

1.17 TF/s

220 TF/s

35.8 TF/s

59.7 GF/s

134 GF/s

0.4 GF/s

Jun-93

Nov-93

Jun-94

Nov-94

Jun-9

5Nov

-95Jun

-96Nov-

96Jun

-97Nov

-97Jun

-98Nov-

98Jun

-99Nov

-99Ju

n-00Nov

-00Jun

-01Nov-

01Jun

-02

Fujitsu'NWT' NAL

NECEarth Simulator

Intel ASCI RedSandia

IBM ASCI WhiteLLNL

N=1

N=500

SUM

1 Gflop/s

1 Tflop/s

100 Mflop/s

100 Gflop/s

100 Tflop/s

10 Gflop/s

10 Tflop/s

1 Pflop/s

My Laptop

7

13

Performance ExtrapolationPerformance Extrapolation

ASCI Purple

Earth Simulator

Jun-9

3Ju

n-94Jun

-95Ju

n-96Ju

n-97Ju

n-98Ju

n-99Jun

-00Ju

n-01Ju

n-02Ju

n-03Jun

-04Ju

n-05Ju

n-06Jun

-07Ju

n-08Jun

-09Ju

n-10

N=1

N=500

Sum

1 GFlop/s

1 TFlop/s

1 PFlop/s

100 MFlop/s

100 GFlop/s

100 TFlop/s

10 GFlop/s

10 TFlop/s

10 PFlop/s

My Laptop

14

ManufacturersManufacturers

HP 168 (12 < 100), IBM 164 (47 < 100)

Cray

SGI

IBM

Sun

HP

TMC

Intel

FujitsuNEC

Hitachiothers

0

100

200

300

400

500

Jun-9

3Nov

-93Ju

n-94Nov

-94Ju

n-95Nov

-95Ju

n-96Nov

-96Ju

n-97Nov

-97Ju

n-98Nov

-98Ju

n-99Nov

-99Ju

n-00Nov

-00Ju

n-01Nov

-01Ju

n-02

8

15

ArchitecturesArchitectures

Single Processor

SMP

MPP

SIMD

Constellation

Cluster - NOW

0

100

200

300

400

500

Jun-93Nov-

93Jun

-94Nov-9

4Jun-95

Nov-95

Jun-96

Nov-96Jun-97

Nov-97

Jun-98

Nov-98

Jun-99Nov-9

9Jun-00

Nov-00

Jun-01Nov-0

1Jun-02

Y-MP C90

Sun HPC

Paragon

CM5T3D

T3E

SP2

Cluster ofSun HPC

ASCI Red

CM2

VP500

SX3

Constellation: # of p/n � n

16

KflopsKflops per Inhabitantper Inhabitant

450

358

245207 203

158141

67

643

0

100

200

300

400

500

600

700

Japan USA

German

y

Scandinav

ia UK

Franc

e

Switzerla

nd Italy

Luxembourg

283

76

White is ES contribution and Blue is ASCI contribution

9

17

80 Clusters on the Top50080 Clusters on the Top500

♦ A total of 42 Intel based and 8 AMD based PC clusters are in the TOP500. Ø31 of these Intel based cluster are IBM

Netfinity systems delivered by IBM. ♦ A substantial part of these are installed

at industrial customers especially in the oil-industry. ØIncluding 5 Sun and 5 Alpha based clusters

and 21 HP AlphaServer. ♦ 14 of these clusters are labeled as

'Self-Made'.

18

Cluster on the Top500Cluster on the Top500

0

10

20

30

40

50

60

70

80

Jun-9

7Nov

-97

Jun-9

8Nov

-98

Jun-9

9Nov

-99Ju

n-00

Nov-00

Jun-0

1Nov

-01Ju

n-02

AMD

Intel

IBM Netfinity

Alpha

HP Alpha Server

Sparc

Processor Breakdown

Pentium III, 37, 46%

Itanium, 2, 3%

Sparc, 5, 6% AMD, 8, 10%

Alpha, 25, 31% Pentium 4, 3,

4%

10

19

Distributed and Parallel SystemsDistributed and Parallel Systems

Distributedsystemshetero-geneous

Massivelyparallelsystemshomo-geneous

Grid

bas

ed

Com

putin

gG

oogle

Netw

ork o

f ws

Clus

ters

w/

spec

ial in

terc

onne

ct

Entro

pia/U

D

Earth

Sim

ulato

r

♦ Gather (unused) resources♦ Steal cycles♦ System SW manages resources♦ System SW adds value♦ 10% - 20% overhead is OK♦ Resources drive applications♦ Time to completion is not critical♦ Time-shared♦ SETI@home

Ø ~ 400,000 machinesØ Averaging 40 Tflop/s

♦ Bounded set of resources ♦ Apps grow to consume all cycles♦ Application manages resources♦ System SW gets in the way♦ 5% overhead is maximum♦ Apps drive purchase of equipment♦ Real-time constraints♦ Space-shared♦ Earth Simulator

Ø 5000 processorsØ Averaging 35 Tflop/s

SETI

@ho

me

Para

llel D

ist

mem

ASCI

Tflo

p/s

20

SETI@homeSETI@home: Global Distributed Computing: Global Distributed Computing♦ Running on 500,000 PCs, ~1000 CPU

Years per DayØ 485,821 CPU Years so far

♦ Sophisticated Data & Signal Processing Analysis

♦ Distributes Datasets from AreciboRadio Telescope

11

21

SETI@homeSETI@home♦ Use thousands of Internet-

connected PCs to help in the search for extraterrestrial intelligence.

♦ When their computer is idle or being wasted this software will download a 300 kilobyte chunk of data for analysis. Performs about 3 Tflops for each client in 15 hours.

♦ The results of this analysis are sent back to the SETI team, combined with thousands of other participants.

♦ Largest distributed computation project in existenceØ2500 machines todayØAveraging 40 Tflop/s

♦ Today a number of companies trying this for profit.

22

Grid Computing Grid Computing --from ET from ET toAnthraxtoAnthrax

12

23

♦ Google query attributesØ150M queries/day (2000/second)Ø3B documents in the index

♦ Data centersØ15,000 Linux systems in 6 data centers

Ø15 TFlop/s and 1000 TB total capabilityØ40-80 1U/2U servers/cabinet Ø100 MB Ethernet switches/cabinet with gigabit

Ethernet uplinkØgrowth from 4,000 systems (June 2000)

Ø18M queries then♦ Performance and operationØ simple reissue of failed commands to new serversØ no performance debugging

Ø problems are not reproducibleSource: Monika Henzinger, Google

24

♦ Today there is a complex interplay and increasing interdependence among the sciences.

♦ Many science and engineering problems require widely dispersed resources be operated as systems.

♦ What we do as collaborative infrastructure developers will have profound influence on the future of science.

♦ Networking, distributed computing, and parallel computation research have matured to make it possible for distributed systems to support high-performance applications, but...Ø Resources are dispersedØ Connectivity is variableØ Dedicated access may not be possible

In the past: IsolationMotivation for Grid ComputingMotivation for Grid Computing

Today: Collaboration

13

25

The Grid

26

IPG NASA http://nas.nasa.gov/~wej/home/IPGGlobus http://www.globus.org/Legion http://www.cs.virgina.edu/~grimshaw/AppLeS http://www-cse.ucsd.edu/groups/hpcl/NetSolve http://www.cs.utk.edu/netsolve/NINF http://phase.etl.go.jp/ninf/Condor http://www.cs.wisc.edu/condor/CUMULVS http://www.epm.ornl.gov/cs/WebFlow http://www.npac.syr.edu/users/gcf/NGC http://www.nordicgrid.net

Grids are HotGrids are Hot

14

27

University of Tennessee Deployment: University of Tennessee Deployment: SScalable calable InIntracampustracampus RResearch esearch GGrid: rid: SInRGSInRG

♦ Federated Ownership: CS, ChemEng., Medical School, Computational Ecology, El. Eng.

♦ Real applications, middleware development, logistical networking

The Knoxville Campus has two DS-3 commodity Internet connections and one DS-3 Internet2/Abilene connection. An OC-3 ATM link routes IP traffic between the Knoxville campus, National Transportation Research Center, andOak Ridge National Laboratory. UT participates in several national networking initiatives including Internet2 (I2),Abilene, the federal Next Generation Internet (NGI) initiative, Southern Universities Research Association (SURA)Regional Information Infrastructure (RII), and Southern Crossroads (SoX).

The UT campus consists of a meshed ATM OC-12 being migrated over to switched Gigabit by early 2002.

28

Grids vs. Capability ComputingGrids vs. Capability Computing♦ Not an “either/or” questionØeach addresses different needsØboth are part of an integrated solution

♦ Grid strengthsØ coupling necessarily distributed resources

Ø instruments, archives, and peopleØeliminating time and space barriers

Øremote resource access and capacity computingØGrids are not a cheap substitute for capability

HPC♦ Capability computing strengthsØ supporting foundational computations

Øterascale and petascale “nation scale” problemsØengaging tightly coupled teams and computations

15

29

Software Technology & PerformanceSoftware Technology & Performance

♦ Tendency to focus on hardware♦ Software required to bridge an ever widening gap♦ Gaps between usable and deliverable performance

is very steepØ Performance only if the data and controls are setup just

rightØOtherwise, dramatic performance degradations, very

unstable situationØWill become more unstable

♦ Challenge of Libraries, PSEs and Tools is formidable with Tflop/s level, even greater with Pflops, some might say insurmountable.

30

Where Does the Performance Go? orWhere Does the Performance Go? orWhy Should I Care About the Memory Hierarchy?Why Should I Care About the Memory Hierarchy?

µProc60%/yr.(2X/1.5yr)

DRAM9%/yr.(2X/10 yrs)1

10

100

1000

1980

1981

1983

1984

1985

1986

1987

1988

1989

1990

1991

1992

1993

1994

1995

1996

1997

1998

1999

2000

DRAM

CPU

1982

Per

form

ance

Time

“Moore’s Law”

Processor-DRAM Memory Gap (latency)

Processor-MemoryPerformance Gap:(grows 50% / year)

16

31

Optimizing Computation and Optimizing Computation and Memory UseMemory Use

♦ Computational optimizationsØTheoretical peak:(# fpus)*(flops/cycle) * Mhz

Ø Pentium 4: (1 fpu)*(2 flops/cycle)*(2.53 Ghz) = 5060 MFLOP/s

♦ Operations like:Ø α = xTy : 2 operands (16 Bytes) needed for 2 flops;

at 5060 Mflop/s will requires 5060 MW/s bandwidthØ y = α x + y : 3 operands (24 Bytes) needed for 2 flops;

at 5060 Mflop/s will requires 7590 MW/s bandwidth

♦ Memory optimizationØTheoretical peak: (bus width) * (bus speed)Ø Pentium 4: (32 bits)*(533 Mhz) = 2132 MB/s = 266 MW/s

32

Memory HierarchyMemory Hierarchy♦ By taking advantage of the principle of locality:Ø Present the user with as much memory as is available in

the cheapest technology.Ø Provide access at the speed offered by the fastest

technology.

Control

Datapath

SecondaryStorage(Disk)

Processor

Registers

MainMemory(DRAM)

Level2 and 3Cache

(SRAM)

On-C

hipC

ache

1s 10,000,000s (10s ms)

100,000 s(.1s ms)

Speed (ns): 10s 100s

100s

Gs

Size (bytes):Ks Ms

TertiaryStorage

(Disk/Tape)

10,000,000,000s (10s sec)

10,000,000 s(10s ms)

Ts

DistributedMemory

Remote ClusterMemory

17

33

Self Adapting SoftwareSelf Adapting Software♦ Software system that …ØObtains information on the underlying system

where they will run.ØAdapts application to the presented data and the

available resources perhaps provide automatic algorithm selectionØDuring execution perform optimization and perhaps

reconfigure based on newly available resources.ØAllow the user to provide for faults and recover

without additional users involvement♦ The moral of the storyØWe know the concepts of how to improve things.ØCapture insights/experience – do what humans do

wellØAutomate the dull stuff

34

Software Generation Software Generation Strategy Strategy -- ATLAS BLASATLAS BLAS

♦ Takes ~ 20 minutes to run, generates Level 1,2, & 3 BLAS

♦ “New” model of high performance programming where critical code is machine generated using parameter optimization.

♦ Designed for modern architecturesØ Need reasonable C compiler

♦ Today ATLAS in used within various ASCI and SciDACactivities and by Matlab, Mathematica, Octave, Maple, Debian, Scyld Beowulf, SuSE,…

♦ Parameter study of the hw ♦ Generate multiple versions

of code, w/difference values of key performance parameters

♦ Run and measure the performance for various versions

♦ Pick best and generate library

♦ Level 1 cache multiply optimizes for:Ø TLB accessØ L1 cache reuseØ FP unit usageØ Memory fetchØ Register reuseØ Loop overhead minimization

18

35

ATLAS ATLAS (DGEMM n = 500)(DGEMM n = 500)

♦ ATLAS is faster than all other portable BLAS implementations and it is comparable with machine-specific libraries provided by the vendor.

♦ Looking at sparse operations

0.0

500.0

1000.0

1500.0

2000.0

2500.0

3000.0

3500.0

AMD Athlon

-600

DEC ev

56-533

DEC ev

6-500

HP9000

/735/1

35

IBM PPC604

-112

IBM Power2

-160

IBM Power3

-200

Intel P-

III 933

MHz

Intel P-

4 2.53

GHz w/SS

E2

SGI R

10000i

p28-20

0

SGI R

12000i

p30-2

70

Sun Ultra

Sparc2

-200

Architectures

MF

LO

P/S

Vendor BLASATLAS BLASF77 BLAS

36

GrADS GrADS -- Grid Application Development Grid Application Development SystemSystem

♦ Problem: Grid has distributed, heterogeneous, dynamic resources; how do we use them?

♦ Goal: reliable performance on dynamically changing resources

♦ Minimize work of preparing an application for Grid executionØ Provide generic versions of key components (currently

built in to applications or manually done)Ø E.g., scheduling, application launch, performance

monitoring

♦ Provide high-level programming tools to help automate application preparationØ Performance modeler, mapper, binder

19

37

NSF/NGS NSF/NGS GrADS GrADS -- GrADSoftGrADSoft ArchitectureArchitecture


Whole-ProgramCompiler

LibrariesBinder

Real-timePerformance

Monitor

PerformanceProblem

ResourceNegotiator

Scheduler

GridRuntimeSystem

SourceAppli-cation

Config-urableObject

Program

SoftwareComponents

Performance Feedback

Negotiation

PIs: Ken Kennedy, Fran Berman, Andrew Chein, Keith Cooper, JD, Ian Foster, Lennart Johnsson, Dan Reed, Carl Kesselman, John Mellor-Crummey, Linda Torczon & Rich Wolski

38

NSF/NGS NSF/NGS GrADS GrADS -- GrADSoftGrADSoft ArchitectureArchitecture


Whole-ProgramCompiler

LibrariesBinder

Real-timePerformance

Monitor

PerformanceProblem

ResourceNegotiator

Scheduler

GridRuntimeSystem

SourceAppli-cation

Config-urableObject

Program

SoftwareComponents

Performance Feedback

Negotiation

PIs: Ken Kennedy, Fran Berman, Andrew Chein, Keith Cooper, JD, Ian Foster, Lennart Johnsson, Dan Reed, Carl Kesselman, John Mellor-Crummey, Linda Torczon & Rich Wolski

20

39

Intelligent ComponentIntelligent Component♦ System to mediate between

user application and multiple possible libraries

♦ Self-Adaptivity and Learning Behavior Ø Heuristics are tuned based on

dataØ System gradually gets smarter

(database)Ø The system can educate the

user

♦ User Interaction Ø User can guide the system by

providing further informationØ System teaches user about

properties of the data

40

Research AreasResearch Areas♦ Automatically generating performance models

(e.g. for ScaLAPACK) on Grid resources♦ Evaluating Performance “Contracts” ♦ Near Optimal Scheduling (execution) on the

Grid♦ Rescheduling for changing resources♦ Checkpointing and fault tolerance♦ High-latency tolerant algorithms (SANS ideas)♦ Porting applications/libraries to GrADS

framework♦ Developing generic GrADSoft interfaces (API’s)

21

41

LAPACK For ClustersLAPACK For Clusters♦ Developing middleware which couples cluster system

information with the specifics of a user problem to launch cluster based applications on the “best” set of resource available.

♦ Using ScaLAPACK as the prototype software, but developing a framework

~ Mbit Switch, (fully connected)

~ Gbit Switch, (fully connected)

Remote memory server, e.g. IBP (TCP/IP)

Local network file server, SUN’s NFS (UDP/IP)e.g. 100 Mbit

Users, etc.

42

GrADS ScaLAPACK versus Non-GrADS ScaLAPACK

0 500 1000 1500 2000 2500

N=20000 - GrADS (14 proc)

N=20000 - Non GrADS (17 proc)





N=5000 - Grads (6 proc)

N=5000 - NonGrADS (17 proc)

Time Taken (seconds)

Grid consists of 17 machines from two heterogeneous, shared (possibly loaded) clusters. GrADS schedules execution on appropriate machines. Non-GraDS uses the ALL the machines.

ScaLAPACK Application Grid Overhead / Spawn GrADS Overhead

22

43

Research DirectionsResearch Directions♦ Parameterizable libraries♦ Fault tolerant algorithms♦ Annotated libraries♦ Hierarchical algorithm libraries♦ “Grid” (network) enabled strategies

A new division of labor between compiler writers, library writers, and algorithm developers and application developers will emerge.

44

Futures for Numerical Algorithms and SoftwareFutures for Numerical Algorithms and Software

♦ Numerical software will be adaptive, exploratory, and intelligent

♦ Determinism in numerical computing will be gone.Ø After all, its not reasonable to ask for exactness in numerical

computations.

ØAuditability of the computation, reproducibility at a cost

♦ Importance of floating point arithmetic will be undiminished.Ø16, 32, 64, 128 bits and beyond.

♦ Reproducibility, fault tolerance, and auditability

♦ Adaptivity is a key so applications can effectively use the resources.

23

45

Collaborators / SupportCollaborators / Support

♦ TOP500ØØH. Mauer, Mannheim UH. Mauer, Mannheim UØØH. Simon, NERSCH. Simon, NERSCØØE. Strohmaier, NERSCE. Strohmaier, NERSC

♦ GrADSØSathish Vadhiyar, UTKØAsim YarKhan, UTKØKen Kennedy, Fran

Berman, Andrew Chein, Keith Cooper, Ian Foster, Carl Kesselman, Lennart Johnsson, Dan Reed, Linda Torczon, & Rich Wolski

ØThanks

NSF Next Generation Software (NGS)

Scientific Discovery through Advanced Computing (SciDAC)

High Performance Computing, Computational Grid, and ...

Documents

Transcript of High Performance Computing, Computational Grid, and ...