Super Computers Parallel Computers_2

8/3/2019 Super Computers Parallel Computers_2

1/59

DONE BY

SREEKANTH.PARJUN.V

AKSHAY.K

ARAVIND.R

SHARATH.S


2/59

By 1960, at the age of 34,

Seymour had established his

reputation for genius in

designing high performancecomputers. He had

completed the design of the

Control Data 1604, the first

computer to be fully

transistorized and had begunthe design of the first system

that earned the title of

supercomputer, the CDC

6600 which was also the firstmajor system to employ

three-dimensional packaging

and an instruction set that

was later to be referred to as

RISC.


3/59


4/59


5/59

Even as a child, Seymour was a problem solver. His sister

tells the story about when Seymour was a young boy, he

rigged a Morse Code connection between his bedroom and

his sister's so that they could communicate after lights out.His father became aware of the late night clicking and told

Seymour to shut down the system because it was bothering

the rest of the household. Seymour's solution was to convert

the clickers to lights and to continue to communicate with his

sister.


6/59


7/59


8/59


9/59


10/59

Robert Frost's, "The Road Not Taken"

"I shall be telling this with a sigh

Somewhere ages and ages hence:

Two roads diverged in a wood, andI--

I took the one less traveled by,

And that has made all thedifference."


11/59

Seymour liked to work with fundamental and simple tools.

Generally only a piece of paper and a pencil. But he

admitted that some of his work required more sophisticated

tools. Once when told that Apple Computer bought a

CRAY to simulate their next Apple computer design,Seymour remarked, "Funny, I am using an Apple to

simulate the CRAY-3." His selection of people for his

projects also reflected fundamentals. Once asked why he

often hires new graduates to help him with early R&D

work, he replied, "Because they don't know that what I'm

asking them to do is impossible, so they try."


12/59

Since the first supercomputer, the Cray-1,was installed at Los Alamos NationalLaboratory in 1976, computational speedhas leaped 500,000 times.

The Cray-1 was capable of 80 megaflops (80million operations a second). The BlueGene/L machine that will be completed nextyear will be five million times faster.


13/59

1: Earth Simulator Center, Japan 2: Intel Itanium2Tiger4 1.4GHz, Quadrics 3: ASCI Q - AlphaServerSC45, 1.25 GHz 4: Blue Gene/L DD1 Prototype(0.5GHz PowerPC 440 w/Custom) 5: PowerEdge1750, P4 Xeon 3.06 GHz, Myrinet 6: eServer pSeries690 (1.9 GHz Power4+) 7: Riken Super CombinedCluster 8: Blue Gene/L DD2 Prototype (0.7 GHzPowerPC 440) 9: Integrity rx2600 Itanium2 1.5 GHz,Quadrics 10: Dawning 4000A, Opteron 2.2 GHz,

Myrinet

June-2004


14/59

November-2004


15/59

Its peak theoretical performance is expected tobe 360 teraflops, and will fit into 64 full racks. It

will also cut down on the amount of heatgenerated by the massive power, a big problemfor supercomputers.

The final machine will help scientists work outthe safety, security and reliability requirementsfor the US's nuclear weapons stockpile, withoutthe need for underground nuclear testing.

IBM's senior vice president of technology andmanufacturing, Nick Donofrio, believes that by2006, Blue Gene will be capable of petaflopcomputing.

This means it would be capable of doing 1,000trillion operations a second.


16/59


17/59

NASA to build 10,000-processor Linuxcomputer

IDG News Service 7/28/04

The National Aeronautics and SpaceAdministration (NASA) has given the green

light to a project that will build the largest eversupercomputer based on Silicon GraphicsInc.'s (SGI) 512-processor Altix computers.

Called Project Columbia, the 10,240-processor

system will be used by researchers at theAdvanced Supercomputing Facility at NASA'sAmes Research Center in Moffett Field,California.


18/59

Scientists will use Columbia to design equipment,simulate future space missions and model weatherpatterns. A portion of the US$160 million system will

also be made available to other government agenciesand educational facilities, said Bill Thigpen, managerof Project Columbia. "We need to look at workingwith other agencies to provide them with access tothis system because it is a unique system," he said.

What makes Project Columbia unique is the size ofthe multiprocessor Linux systems, or nodes, that itclusters together. It is common for supercomputersto be built of thousands of two-processor nodes, butthe Ames system uses SGI's NUMAlink switchingtechnology and ProPack Linux operating systemenhancements to connect 512-processor nodes, eachof which will have more than 1,000G bytes of

memory


19/59

"We use a very large single-system image,"

said Jeff Greenwald, senior director of serverproduct marketing with SGI. "The other guyscome with a very thin node cluster, and try toscrew them all together."

The Altix nodes will use Intel Corp.'s Itanium2 microprocessors, and the entire 20-nodesystem is expected to be fully assembled byyear's end, he said.

SGI has used this large-node technology tobuild a number of smaller Altix systems withbetween 3,000 and 6,000 processors, butProject Columbia will be the largest to date,Greenwald said


20/59


21/59

The Earth Simulator has held on to thetop spot since June 2002. It is dedicated

to climate modelling and simulatingseismic activity


22/59


23/59


24/59

SINGAPORE (CNN) -- A group of researchers

from Singapore has created a computer chip

that has the power of 100 standard

computers.

The group of five, all working at Ngee AnnPolytechnic, will commercialize their development

by January and sell it to the pharmaceuticalindustry, where they say the invention will savetime and money.

Lead researcher Darran Nathan, 24, explains that

unlike standard computer chips, which functionusing software, his is based on a computer'shardware.


25/59


26/59

"An ordinary computer chip will interpret instructionsfrom the software and execute a command," he says.

"Our chip is a reconfigurable chip, which means itdownloads an actual file to the chip and rewires itaccording to subsequent processing done in thehardware."

Nathan says the process is highly technical but, putsimply, is a computer chip that works at a speed of100 standard computers combined.

He says the super chip was originally created withthe telecommunications industry in mind, but soonafter work on the project began two years ago, theyrealized the benefits would be much more useful tolife sciences.


27/59

"It is 100 times quicker than your standard computer.Most people do not need such a powerful computer, but

in the area of designing and developing drugs, it ishugely important," says Nathan.

"It basically means getting essential drugs on the streetquicker, at a cheaper cost."

Nathan says the device will cost between US$30,000and US$61,000, and its key point of difference betweenother supercomputers is its small size.

The team, which calls itself Project Proteus, after theshape-shifting Greek god, are aged between 24 and 27.

Last week they showcased their chip at the GlobalEntrepolis convention in Singapore where Mr Nathansays they received a lot of positive feedback.


28/59

A Supercomputer at $5.2 million

Virginia Tech 1,100node Macs.

G5supercomputer

A Supercomputer at $5.2 million


29/59

The Virginia Polytechnic Institute and State

University has built a supercomputercomprised of a cluster of 1,100 dual-processor Macintosh G5 computers. Basedon preliminary benchmarks, Big Mac iscapable of 8.1 teraflops per second. The

Mac supercomputer still is being fine tuned,and the full extent of its computing powerwill not be known until November. But the8.1 teraflops figure would make the Big

Mac the world's fourth fastestsupercomputer


30/59

Big Mac's cost relative to similarmachines is as noteworthy as itsperformance. The Apple supercomputerwas constructed for just over US$5million, and the cluster was assembled in

about four weeks.In contrast, the world's leadingsupercomputers cost well over $100million to build and require several years

to construct. The Earth Simulator, whichclocked in at 38.5 teraflops in 2002,reportedly cost up to $250 million.


31/59

Srinidhi Varadarajan, Ph.D.Dr. Srinidhi Varadarajan is an Assistant

ProfessorofComputer Science atVirginiaTech. He washonored withtheNSF Career

Award in2002for "Weaving a CodeTapestry: A CompilerDirected Frameworkfor ScalableNetwork Emulation." Hehasfocusedhisresearchonbuilding adistributednetworkemulationsystemthatcanscaletoemulatehundredsofthousandsofvirtual nodes.

October28Time: 7:30pm- 9:00pm

Location: Santa Clara Ballroom


32/59

Parallel Computersy Two common types

y Cluster

y

Multi-Processor


33/59

Cluster Computers


34/59

Clusters on the Rise

Using clusters of small machines to build asupercomputer is not a new concept.

Another of the world's top machines, housedat the Lawrence Livermore NationalLaboratory, was constructed from 2,304Xeon processors. The machine was build by

Utah-based Linux Networx.

Clustering technology has meant thattraditional big-iron leaders like Cray (Nasdaq:CRAY) and IBM have new competition from

makers of smaller machines. Dell (Nasdaq:DELL) , among other companies, has soldhigh-powered computing clusters to researchinstitutions.


35/59

Cluster Computersy Each computer in a cluster is a complete computer by

itselfy CPU

y Memory

y Disk

y etc

y Computers communicate with each other via some

interconnection bus


36/59

Cluster Computersy Typically used where one computer does not have

enough capacity to do the expected work

y

Large Serversy Cheaper than building one GIANT computer


37/59

Although not new, supercomputing clusteringtechnology still is impressive. It works byfarming out chunks of data to individual

machines, adding that clustering works betterfor some types of computing problems thanothers.

For example, a cluster would not be ideal to

compete against IBM's Deep Bluesupercomputer in a chess match; in this case,all the data must be available to oneprocessor at the same moment -- themachine operates much in the same way asthe human brain handles tasks.

However, a cluster would be ideal for theprocessing of seismic data for oil exploration,because that computing job can be dividedinto many smaller tasks.


38/59

Cluster Computersy Need to break up work among the computers in the

cluster

y

Example: Microsoft.com Search Enginey 6 computers running SQL Server

y Each has a copy of the MS Knowledge Base

y Search requests come to one computer

y

Sends request to one of the 6y Attempts to keep all 6 busy


39/59

The Virginia Tech Mac supercomputershould be fully functional and in use byJanuary 2004. It will be used for research

into nanoscale electronics, quantumchemistry, computational chemistry,aerodynamics, molecular statics,computational acoustics and the molecular

modeling of proteins.


40/59

Multiprocessors

I/O PortDevice

DeviceController

CPU

Bus

Memory

CPU

CPU


41/59

Multiprocessorsy Systems designed to have 2 to 8 CPUs

y The CPUs all share the other parts of the computer

y Memoryy Disk

y System Bus

y etc

y CPUs communicate via Memory and the SystemBus


42/59

MultiProcessorsy Each CPU shares memory, disks, etc

y Cheaper than clusters

y

Not as good performance as clustersy Often used for

y Small Servers

y High-end Workstations


43/59

MultiProcessorsy OS automatically shares work among available CPUs

y On a workstation

y

One CPU can be running an engineering design programy Another CPU can be doing complex graphics formatting


44/59

Specialized ProcessorsyVector Processors

y Massively Parallel Computers


45/59

Vector Processors

For (I=0;I


46/59

Vector ProcessorsSpecial instructions to operate on vectors (arrays)

y Vector instruction specifiesy

Starting addresses of all 3 arraysy Loop count

y Saves For Loop overhead

y Can more efficiently access memory

y

Also Known as SIMD Computersy Single Instruction Multiple Data


47/59

Vector Processorsy Until the 1990s, the worlds fastest supercomputers

were implemented as vector processors

y

Now, Vector Processors are typically special peripheraldevices that can be installed on a regular computer


48/59

Massively Parallel Computersy IBM ASCI Purple

y Cluster of 196 computers

y

Each computer hasy 64 CPUs

y 256 Gigabytes of RAM

y 10,000 GB of Disk


49/59

Massively Parallel Computery How will ASCI Purple be used?

y Simulation of molecular dynamics

y

Research into repairing damaged DNAy Analysis of seismic waves

y Earthquake research

y Simulation of star evolution

y

Simulation of Weapons of Mass Destruction


50/59

Acco

rdingt

othe article, thesuperc

omputer,poweredby2,200 IBM G5processors, has

been initiallyrated atcomputing7.41 trillionoperationspersecond. Thefinal numbercouldbemuchhigher, accordingtoschool officials,

but ifnot, itwouldrank asthe #4fastestsupercomputingcluster inthe world.

Japan's US$250M Earth Simulator, which is

currentlythe wo

rld

'sfastestc

omputer

LawrenceLivermore's US$10-15M clustersystem,which ismadeupof2,304 Intel Xeonprocessors.IBM recently installed "Pacific Blue" attheLawrenceLivermoreLaboratoriesfor $94million


51/59

"We aredemonstratingthatyoucanbuild averyhighperformancemachinefor a fifthto a

tenthofthecostofwhatsupercomputersnowcost," said Hassan Aref, thedeanofthe SchoolofEngineering atVirginia Tech in Blacksburg

1998 a groupcalleddistributed.net linkedthousandsof

computersofall kinds aroundthe worldvia the Internet, andcracked a 56-bitDES-II code in40days. Ithadpreviouslybeenthoughtthatsuchheavyweightciphers wouldtakehundredsofyearstocrackevenonfastcomputers. OneversionoftheDistributed.netprogramran as a screensaverthatkicked in, andbegancrackingcode, wheneverthemachine was idleformorethan a few minutes.Distributed.netbills itselfasthe "FastestComputeron Earth",eventhoughtheirhardwarebill iseffectively zero.


52/59

The idea isstraightforward. Yousetup an arbitrary

numberofPCs, networkthem, typicallyusingfastEthernet, andthensendthemproblemsthatcanbedividedup amongthemachines'processors. Onemachine acts as a serverthatsyncsup all therest,calledclients.

Beowulfspecssoftware likethe Message PassingInterface writtenundertheLinuxoperatingsystem,thatallowsthemachinestocommunicate whileworkingontheproblem.

AndsinceLinux, brainchildofcomputersciencestudentLinus Torvalds, isfree, itkeepsthecostdown


53/59

Modelingthetrajectoriesoftensof

millionsofchargedparticles, eachinteracting withtheothersthroughelectro-magneticforces, requiresheavy-dutynumbercrunching. Toharnesssupercomputingpower ata

desktopprice, UCLAsDr. Viktor K.Decyk andhiscolleagueshavecreatedtheirownsuper-fast,parallel processingsupercomputer

using a clusterof

Power Macint

oshcomputers.


54/59

Apple's G4 Cubes used for cell

mutation detection and

genotyping analysisSYDNEY - 22 January

2001


55/59

World's fastest" Macintosh cluster

Researchers attheGrupodeLaserse Plasmas(GoLP) in Portugal havecreated whattheybill

asthe world'sfastestMacintosh-basedcluster.Consistingof16dual-processor Power MacG4/450s, theclusterdeliversmorethan50GigaFlopsofpeakpower andtook justoneday

tosetup.


56/59

Apple Computerpurchased a big Craysupercomputerinthemid-1980s. Infact, SteveJobs was Cray'sfirst

andonly walk-incustomer. He arrivedunannounced(sothestorygoes) atCrayheadquarters in MendotaHeights, Minnesota and askedtospeaktosomeoneaboutbuying a Cray. Theynearlythrew himout. It'sonlyslightly lesseccentricthansomeone walking into

NASA Johnson Space Center and inquiringhow topurchase a shuttleorbiter.

Later, CraypresidentJohnRollwagenphonedSeymour andtoldhimthatApplehad justpurchased a

Craythatwouldbeused indesigningthenextMacintosh. Seymourthoughtfor a bit, andrepliedthatthatseemedreasonable, sincehe wasusing aMacintoshtodesignthenextCray!


57/59


58/59


59/59

Super Computers Parallel Computers_2

Documents

Super Computers Parallel Computers_2