08/27/2002 CS267-Lecture 1 1

CS267Applications of Parallel

Computers

Lecture 1: Introduction

Horst D. Simon

hdsimon@lbl.gov

http://www.nersc.gov/~simon

08/27/2002 CS267-Lecture 1 2

Outline

• Introduction

• Large important problems require powerful computers

• Why powerful computers must be parallel processors

• Principles of parallel computing performance

• Structure of the course

08/27/2002 CS267-Lecture 1 3

Why we need powerful computers

08/27/2002 CS267-Lecture 1 4

Simulation: The Third Pillar of Science

• Traditional scientific and engineering paradigm:1) Do theory or paper design.

2) Perform experiments or build system.

• Limitations:- Too difficult -- build large wind tunnels.

- Too expensive -- build a throw-away passenger jet.

- Too slow -- wait for climate or galactic evolution.

- Too dangerous -- weapons, drug design, climate experimentation.

• Computational science paradigm:3) Use high performance computer systems to simulate the

phenomenon- Base on known physical laws and efficient numerical methods.

08/27/2002 CS267-Lecture 1 5

Some Particularly Challenging Computations• Science

- Global climate modeling- Astrophysical modeling- Biology: genomics; protein folding; drug design- Computational Chemistry- Computational Material Sciences and Nanosciences

• Engineering- Crash simulation- Semiconductor design- Earthquake and structural modeling- Computational fluid dynamics- Combustion

• Business- Financial and economic modeling- Transaction processing, web services and search engines

• Defense- Nuclear weapons -- test by simulations- Cryptography

08/27/2002 CS267-Lecture 1 6

Units of Measure in HPC

• High Performance Computing (HPC) units are:- Flop/s: floating point operations

- Bytes: size of data

• Typical sizes are millions, billions, trillions…Mega Mflop/s = 106 flop/sec Mbyte = 106 byte

(also 220 = 1048576)

Giga Gflop/s = 109 flop/sec Gbyte = 109 byte

(also 230 = 1073741824)

Tera Tflop/s = 1012 flop/sec Tbyte = 1012 byte

(also 240 = 10995211627776)

Peta Pflop/s = 1015 flop/sec Pbyte = 1015 byte

(also 250 = 1125899906842624)

Exa Eflop/s = 1018 flop/sec Ebyte = 1018 byte

08/27/2002 CS267-Lecture 1 7

Economic Impact of HPC

• Airlines:- System-wide logistics optimization systems on parallel systems.

- Savings: approx. $100 million per airline per year.

• Automotive design:- Major automotive companies use large systems (500+ CPUs) for:

- CAD-CAM, crash testing, structural integrity and aerodynamics.

- One company has 500+ CPU parallel system.

- Savings: approx. $1 billion per company per year.

• Semiconductor industry:- Semiconductor firms use large systems (500+ CPUs) for

- device electronics simulation and logic validation

- Savings: approx. $1 billion per company per year.

• Securities industry:- Savings: approx. $15 billion per year for U.S. home mortgages.

08/27/2002 CS267-Lecture 1 8

Global Climate Modeling Problem

• Problem is to compute:f(latitude, longitude, elevation, time)

temperature, pressure, humidity, wind velocity

• Approach:- Discretize the domain, e.g., a measurement point every 10 km

- Devise an algorithm to predict weather at time t+1 given t

• Uses:- Predict major events,

e.g., El Nino

- Use in setting air emissions standards

Source: http://www.epm.ornl.gov/chammp/chammp.html

08/27/2002 CS267-Lecture 1 9

Global Climate Modeling Computation

• One piece is modeling the fluid flow in the atmosphere- Solve Navier-Stokes problem

- Roughly 100 Flops per grid point with 1 minute timestep

• Computational requirements:- To match real-time, need 5x 1011 flops in 60 seconds = 8 Gflop/s

- Weather prediction (7 days in 24 hours) 56 Gflop/s

- Climate prediction (50 years in 30 days) 4.8 Tflop/s

- To use in policy negotiations (50 years in 12 hours) 288 Tflop/s

• To double the grid resolution, computation is at least 8x

• State of the art models require integration of atmosphere, ocean, sea-ice, land models, plus possibly carbon cycle, geochemistry and more

• Current models are coarser than this

High Resolution Climate Modeling on NERSC-3 – P. Duffy,

et al., LLNL

08/27/2002 CS267-Lecture 1 11

Comp. Science : A 1000 year climate simulation

• Warren Washington and Jerry Meehl, National Center for Atmospheric Research; Bert Semtner, Naval Postgraduate School; John Weatherly, U.S. Army Cold Regions Research and Engineering Lab Laboratory et al.

• A 1000-year simulation demonstrates the ability of the new Community Climate System Model (CCSM2) to produce a long-term, stable representation of the earth’s climate. •760,000 processor hours usedhttp://www.nersc.gov/aboutnersc/pubs/bigsplash.pdf

08/27/2002 CS267-Lecture 1 12

Comp. Science: High Resolution Global Coupled Ocean/Sea Ice Model

• Mathew E. Maltrud, Los Alamos National Laboratory; Julie L. McClean, Naval Postgraduate School.

• The objective of this project is to couple a high-resolution ocean general circulation model with a high-resolution dynamic-thermodynamic sea ice model in a global context.

•Currently, such simulations are typically performed with a horizontal grid resolution of about 1 degree. This project is running a global ocean circulation model with horizontal resolution of approximately 1/10th degree.•Allows resolution of geographical features critical for climate studies such as Canadian Archipelagohttp://www.nersc.gov/aboutnersc/pubs/bigsplash.pdf

08/27/2002 CS267-Lecture 1 13

Parallel Computing in Web Search

• Functional parallelism: crawling, indexing, sorting

• Parallelism between queries: multiple users

• Finding information amidst junk

• Preprocessing of the web data set to help find information

• General themes of sifting through large, unstructured data sets:

- when to put white socks on sale

- what advertisements should you receive

- finding medical problems in a community

08/27/2002 CS267-Lecture 1 14

Document Retrieval Computation

• Approach:- Store the documents in a large (sparse) matrix

- Use Latent Semantic Indexing (LSI), or related algorithms to “partition”

- Needs large sparse matrix-vector multiply

# keywords

~100K

# documents ~= 10 M

24 65 18

•Matrix is compressed•“Random” memory access•Scatter/gather vs. cache miss per 2Flops

Ten million documents in typical matrix. Web storage increasing 2x every 5 months.Similar ideas may apply to image retrieval.

x

08/27/2002 CS267-Lecture 1 15

Transaction Processing

• Parallelism is natural in relational operators: select, join, etc.

• Many difficult issues: data partitioning, locking, threading.

(mar. 15, 1996)

0

5000

10000

15000

20000

25000

0 20 40 60 80 100 120

Processors

Th

rou

gh

pu

t (t

pm

C)

other

Tandem Himalaya

IBM PowerPC

DEC Alpha

SGI PowerChallenge

HP PA

08/27/2002 CS267-Lecture 1 16

Why powerful computers are

parallel

08/27/2002 CS267-Lecture 1 17

Technology Trends: Microprocessor Capacity

2X transistors/Chip Every 1.5 years

Called “Moore’s Law”

Moore’s Law

Microprocessors have become smaller, denser, and more powerful.

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

Slide source: Jack Dongarra

08/27/2002 CS267-Lecture 1 18

Impact of Device Shrinkage

• What happens when the feature size shrinks by a factor of x ?

• Clock rate goes up by x - actually less than x, because of power consumption

• Transistors per unit area goes up by x2

• Die size also tends to increase- typically another factor of ~x

• Raw computing power of the chip goes up by ~ x4 !- of which x3 is devoted either to parallelism or locality

08/27/2002 CS267-Lecture 1 19

Microprocessor Transistors

Pentium

R10000

R2000R3000

i8086

i8080

i80386

i80286

i4004

1,000

10,000

100,000

1,000,000

10,000,000

100,000,000

1970 1975 1980 1985 1990 1995 2000 2005

Year

Tra

nsis

tors

08/27/2002 CS267-Lecture 1 20

Microprocessor Clock Rate

0.1

1

10

100

1000

1970 1975 1980 1985 1990 1995 2000 2005

Year

Clo

ck

Ra

te (

MH

z)

08/27/2002 CS267-Lecture 1 21

Empirical Trends: Microprocessor Performance

C90

Ymp

Xmp

Xmp

Cray 1s

IBM Power2/990

MIPS R4400

HP9000/735

DEC Alpha AXPHP 9000/750

IBM RS6000/540

MIPS M/2000

MIPS M/120

Sun 4/2601

10

100

1000

10000

1975 1980 1985 1990 1995 2000

Lin

pa

ck

MF

LO

PS

Cray n=1000Cray n=100Micro n=1000Micro n=100

DEC 8200

T94

08/27/2002 CS267-Lecture 1 22

SIA Projections for Microprocessors

Compute power ~1/(Feature Size)3

0.01

0.1

1

10

100

10001

99

5

19

98

20

01

20

04

20

07

20

10

Year of Introduction

Fe

atu

re S

ize

(m

icro

ns

) &

Mill

ion

T

ran

sis

tors

pe

r c

hip

Feature Size(microns)

Transistors perchip x 10**(-6)

based on F.S.Preston, 1997

08/27/2002 CS267-Lecture 1 23

But there are limiting forces: Increased cost and difficulty of manufacturing

• Moore’s 2nd law (Rock’s law)

Demo of 0.06 micron CMOS

08/27/2002 CS267-Lecture 1 24

How fast can a serial computer be?

• Consider the 1 Tflop/s sequential machine:

- Data must travel some distance, r, to get from memory to CPU.

- Go get 1 data element per cycle, this means 1012 times per second at the speed of light, c = 3x108 m/s. Thus r < c/1012 = 0.3 mm.

• Now put 1 Tbyte of storage in a 0.3 mm x 0.3 mm area:

- Each word occupies about 3 square Angstroms, or the size of a small atom.

r = 0.3 mm1 Tflop/s, 1 Tbyte sequential machine

08/27/2002 CS267-Lecture 1 25

Microprocessor Transistors and Parallelism

Pentium

R10000

R2000R3000

i8086

i8080

i80386

i80286

i4004

1,000

10,000

100,000

1,000,000

10,000,000

100,000,000

1970 1975 1980 1985 1990 1995 2000 2005

Year

Tra

nsis

tors

Bit-LevelParallelism

Instruction-LevelParallelism

Thread-LevelParallelism?

08/27/2002 CS267-Lecture 1 26

“Automatic” Parallelism in Modern Machines• Bit level parallelism: within floating point operations, etc.

• Instruction level parallelism (ILP): multiple instructions execute per clock cycle.

• Memory system parallelism: overlap of memory operations with computation.

• OS parallelism: multiple jobs run in parallel on commodity SMPs.

There are limitations to all of these!

Thus to achieve high performance, the programmer needs to identify, schedule and coordinate parallel tasks and data.

08/27/2002 CS267-Lecture 1 27

The Opportunity: Dramatic Advances in ComputingTerascale Today, Petascale Tomorrow

0.1

1

10

100

1,000

1998 2000 2002

Pea

k T

eraf

lop

s

20041996 2006

Increased Use of Parallelism

Microprocessor Advances

MICROPROCESSORS

2x increase in microprocessor speeds every 18-24 months

(“Moore’s Law”)

PARALLELISM

More and more processors being used on single problem

INNOVATIVE DESIGNS

Processors-in-MemoryHTMT

IBM “Blue Gene”Innovative Designs

08/27/2002 CS267-Lecture 1 28

Technology Trends in Parallel Computers

08/27/2002 CS267-Lecture 1 29

• Microprocessors have made desktop computing in 2000 what supercomputing was in 1990.

• Massive Parallelism has changed the “high end” completely.

• Today clusters of Symmetric Multiprocessors are the standard supercomputer architecture.

Nevertheless, the microprocessor revolution will continue with little attenuation for ~10 years.

08/27/2002 CS267-Lecture 1 30

A Parallel Computer Today: NERSC-3 Vital Statistics

• 5 Teraflop/s Peak Performance – 3.05 Teraflop/s with Linpack- 208 nodes, 16 CPUs per node at 1.5 Gflop/s per CPU

- “Worst case” Sustained System Performance measure .358 Tflop/s (7.2%)

- “Best Case” Gordon Bell submission 2.46 on 134 nodes (77%)

• 4.5 TB of main memory- 140 nodes with 16 GB each, 64 nodes with 32 GBs, and 4 nodes with 64 GBs.

• 40 TB total disk space- 20 TB formatted shared, global, parallel, file space; 15 TB local disk for system usage

• Unique 512 way Double/Single switch configuration

08/27/2002 CS267-Lecture 1 31

TOP500 – June 2002 (see www.top500.org)

08/27/2002 CS267-Lecture 1 32

TOP500 - Performance

1.17 TF/s

220 TF/s

35.8 TF/s

59.7 GF/s

134 GF/s

0.4 GF/s

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

08/27/2002 CS267-Lecture 1 33

Manufacturers

Cray

SGI

IBM

Sun

HP

TMC

Intel

FujitsuNEC

Hitachiothers

0

100

200

300

400

500

08/27/2002 CS267-Lecture 1 34

Manufacturers

Cray

SGI

IBM

SunHP

TMC

Intel

Fujitsu

NECHitachiothers

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

Pe

rfo

rma

nc

e

08/27/2002 CS267-Lecture 1 35

Processor Type

Scalar

VectorSIMD

0

100

200

300

400

500

08/27/2002 CS267-Lecture 1 36

Chip Technology

Alpha

Power

HP

intel

MIPS

Sparcother COTS

proprietary

0

100

200

300

400

500

08/27/2002 CS267-Lecture 1 37

Architectures

Single Processor

SMP

MPP

SIMD

Constellation

Cluster - NOW

0

100

200

300

400

500

Y-MP C90

Sun HPC

Paragon

CM5T3D

T3E

SP2

Cluster of Sun HPC

ASCI Red

CM2

VP500

SX3

08/27/2002 CS267-Lecture 1 38

NOW - Cluster

0

10

20

30

40

50

60

70

80AMD

Intel

IBM Netfinity

Alpha

HP Alpha Server

Sparc

Why do we have only commodity components?

Dead Supercomputer Society

• ACRI

• Alliant

• American Supercomputer

• Ametek

• Applied Dynamics

• Astronautics

• BBN

• CDC

• Convex

• Cray Computer

• Cray Research

• Culler-Harris

• Culler Scientific

• Cydrome

• Dana/Ardent/Stellar/Stardent

• Denelcor

• Elexsi

• ETA Systems

• Evans and Sutherland Computer

• Floating Point Systems

• Galaxy YH-1

• Goodyear Aerospace MPP

• Gould NPL

• Guiltech

• Intel Scientific Computers

• International Parallel Machines

• Kendall Square Research

• Key Computer Laboratories

• MasPar

• Meiko

• Multiflow

• Myrias

• Numerix

• Prisma

• Thinking Machines

• Saxpy

• Scientific Computer Systems (SCS)

• Soviet Supercomputers

• Supertek

• Supercomputer Systems

• Suprenum

• Vitesse Electronics

08/27/2002 CS267-Lecture 1 41

Warm Up Homework

Assignment

08/27/2002 CS267-Lecture 1 42

The Parallel Computing Challenge: improving real performance of scientific applications

0.1

1

10

100

1,000

2000 2004T

eraf

lop

s1996

Peak Performance is skyrocketing In 1990’s, peak performance increased 100x; in

2000’s, it will increase 1000x

But ... Efficiency declined from 40-50% on the vector

supercomputers of 1990s to as little as 5-10% on parallel supercomputers of today

Close the gap through ... Mathematical methods and algorithms that

achieve high performance on a single processor and scale to thousands of processors

More efficient programming models for massively parallel supercomputers

Parallel Tools

PerformanceGap

Peak Performance

Real Performance

08/27/2002 CS267-Lecture 1 43

Performance Levels

• Peak advertised performance (PAP)- You can’t possibly compute faster than this speed

• LINPACK (TPP)- The “hello world” program for parallel computing

• Gordon Bell Prize winning applications performance- The right application/algorithm/platform combination plus years of work

• Average sustained applications performance- What one reasonable can expect for standard applications

When reporting performance results, these levels are often confused, even in reviewed publications

08/27/2002 CS267-Lecture 1 44

Performance Levels (for example on NERSC-3)

• Peak advertised performance (PAP): 5 Tflop/s

• LINPACK (TPP): 3.05 Tflop/s

• Gordon Bell Prize winning applications performance : 2.46 Tflop/s

- Material Science application at SC01

• Average sustained applications performance: ~0.4 Tflop/s

- Less than 10% peak!

08/27/2002 CS267-Lecture 1 45

First Assignment• See home page for details.

• Find an application of parallel computing and build a web page describing it.

- Choose something from your research area.- Or from the web or elsewhere.

• Create a web page describing the application. - Describe the application and provide a reference (or link)- Describe the platform where this application was run- Find peak and LINPACK performance for the platform and its rank on

the TOP500 list- Find performance of your selected application- What ratio of sustained to peak performance is reported?- Evaluate project: How did the application scale? What were the major

difficulties in obtaining good performance? What tools and algorithms were used?

• Send us (Horst and David) the link and add the webpage to your portfolio

• Due next week, Thursday (9/5).

08/27/2002 CS267-Lecture 1 46

Course Organization

08/27/2002 CS267-Lecture 1 47

Schedule of Topics• Introduction

• Parallel Programming Models and Machines- Shared Memory and Multithreading

- Distributed Memory and Message Passing

- Data parallelism

• Sources of Parallelism in Simulation

• Tools - Languages (UPC)

- Performance Tools

- Visualization

- Environments

• Algorithms- Dense Linear Algebra

- Partial Differential Equations (PDEs)

- Particle methods

- Load balancing, synchronization techniques

- Sparse matrices

• Applications: biology, climate, combustion, astrophysics

• Project Reports

08/27/2002 CS267-Lecture 1 48

Reading Materials• Some on-line texts:

- Demmel’s notes from CS267 Spring 1999, which are similar to 2000 and 2001. However, they contain links to html notes from 1996.

- http://www.cs.berkeley.edu/~demmel/cs267_Spr99/

- Yelick’s notes from Fall 2001

- http://www.cs.berkeley.edu/~dbindel/cs267ta/

- Ian Foster’s book, “Designing and Building Parallel Programming”.

- http://www-unix.mcs.anl.gov/dbpp/

• Recommended text:- “Sourcebook for Parallel Computing”, by Dongarra, Foster, Fox, ..

- Available in bookstores in November 2002; now available as a reader or on CD;

• Potentially Useful- “Performance Optimization of Numerically Intensive Codes” by Stefan

Goedecker and Adolfy Hoisie

- This is a practical guide to optimization, mostly for those of you who have never done any optimization

08/27/2002 CS267-Lecture 1 49

Other Topics or Interest

• Field Trips- NERSC Visualization lab, Thursday, October 3 confirmed

- Silicon Valley/Computer History Museum, Tuesday, November 26 ?

• Projects- MATLAB anyone? There is a parallel MATLAB available on seaborg

- Student Volunteer at SC2002 in Baltimore, November 16 – 22, 2002

- http://hpc.ncsa.uiuc.edu:8080/sc2002/svol_registration.html for the student volunteer program

- http://www.sc2002.org for the conference

• Also of interest: - ACTS Parallel Tools Workshop for Students in Berkeley, Sept. 4 – 7,

2002 see http://acts.nersc.gov/workshop ; send e-mail to Osni Marques at oamarques@lbl.gov to register; about five spots available for CS267

08/27/2002 CS267-Lecture 1 50

Requirements• Fill out on-line account request for Millennium machine.

- See course web page for pointer

• Fill out request for NERSC account- Form available in class

• Fill out survey- e-mail to David if you missed this lecture

• Build a portfolio- Every week or two students will report explorations, ideas, proposed

work, and work to the TA via an organized webpage, document or notebook.

- There will be about four programming assignments geared towards hands-on experience, interdisciplinary teams.

- There will be a Final Project

- Teams of 2-3, interdisciplinary is best.

- Interesting applications or advance of systems.

- Presentation (poster session)

- Conference quality paper

08/27/2002 CS267-Lecture 1 51

What you should get out of the course

In depth understanding of:

• When is parallel computing useful?

• Understanding of parallel computing hardware options.

• Overview of programming models (software) and tools.

• Some important parallel applications and the algorithms

• Performance analysis and tuning

08/27/2002 CS267-Lecture 1 52

Administrative Information• Instructors:

- Horst D. Simon, 329 Soda, hdsimon@lbl.gov, (510) 486-7377- TA: David Garmire, 566 Soda, strive@eecs.berkeley.edu, (510) 643

6763

• Accounts – fill out online registration for Millenium; fill out form for NERSC accounts on seaborg

• Class survey – fill out today

• Lecture notes are based on previous semester notes:- Jim Demmel, David Culler, David Bailey, Bob Lucas, Kathy Yelick

and myself

• Reader based on “Sourcebook for Parallel Computing”; hardcopy or CD option

• Discussion section: Wed. 1:30 – 2:30 in 405 Soda; not every week

• Most class material and lecture notes are at: - http://www.cs.berkeley.edu/~strive/cs267