Who uses a supercomputer anyway? Jim Greer. Start of Digital Computer: ENIAC Built in 1943-45 at the Moore School of the University of Pennsylvania for.

Who uses a supercomputer anyway?

Jim Greer

Start of Digital Computer: ENIAC

Built in 1943-45 at the Moore School of the University of Pennsylvania for the War effort by John Mauchly and J. Presper Eckert.

The Electronic Numerical Integrator And Computer (ENIAC) was one of the first general-purpose electronic digital computers.

Programming the Computer

Programming in early computers is by wiring the cables and flipping the switches.

MANIAC

The MANIAC had a memory of 1K 40-bit words.

Multiplication took a milli-second.

THE JOURNAL OF CHEMICAL PHYSICS VOLUME 21, NUMBER 6 JUNE, 1953

Equation of State Calculations by Fast Computing Machines

NICHOLAS METROPOLIS, ARIANNA W. ROSENBLUTH, MARSHALL N. ROSENBLUTH, AND AUGUSTA H. TELLER,

Los Alamos Scientific Laboratory, Los Alamos, New Mexico

AND

EDWARD TELLER, * Department of Physics, University of Chicago, Chicago, Illinois

(Received March 6, 1953)

A general method, suitable for fast computing machines, for investigating such properties as equations of state for substances consisting of interacting individual molecules is described. The method consists of a modified Monte Carlo integration over configuration space. Results for the two-dimensional rigid-sphere system have been obtained on the Los Alamos MANIAC and are presented here. These results are compared to the free volume equation of state and to a four-term virial coefficient expansion.

1087This algorithm by Metropolis et al, from1953 has been cited as among the top 10 algorithms having the "greatest influence on the development and practice of science and engineering in the 20th century."

Machine is long out of date-the methods and scientific approach used remain relevant today.

Basis for simulated annealing

Year Computer Name Power (Watts) Performance (adds/sec) Memory (kByte) Price

(US dollars)

1951 UNIVAC I 124,500 1,900 48 $1,000,000

1964 IBM S360 10,000 500,000 64 $1,000,000

1965 PDP-8 500 330,000 4 $16,000

1976 Cray-1 60,000 166,000,000 32,768 $4,000,000

1981 IBM PC 150 240,000 256 $3,000

1991 HP 9000 500 50,000,000 16,384 $7,400

2005 IBM notebook 20 1,000,000,000 512,000 $1,900

What do the machines look like nowadays?

http://news.bbc.co.uk/1/hi/technology/6128066.stm

When HPCx first came into service in 2002 it was one of the top 10 fastest supercomputers in the world. Despite an upgrade in 2004, it has since slipped to 59th place in the Top 500 supercomputers list.

At the moment the most potent machine in the world is the IBM's Blue Gene/L at the Lawrence Livermore National Laboratory in California where it is used to ensure that the US nuclear weapons stockpile remains safe and reliable.

… only machine to have pushed through the 100 teraflop barrier, performs a staggering 280.6 trillion calculations per second …. 367 teraflops

Seymour Cray - plumber

Beowulf is perhaps the most well-known type of parallel processing cluster today. Donald Becker and Thomas Sterling designed the first Beowulf prototype in 1994 for NASA. It consisted of 16 486-DX4 processors connected by channel-bonded Ethernet. The next Beowulf clusters were built around 16 Pentium Pro (P6) 200-MHz processors connected by Fast Ethernet adapters and switches.

Beowulf is a design for high-performance parallel computing clusters on inexpensive personal computer hardware. Originally developed at NASA, Beowulf systems are now

deployed worldwide, chiefly in support of scientific computing.

A Beowulf cluster is a group of usually identical PC computers running a Free and Open Source Software (FOSS) Unix-like operating system, such as Linux or BSD. They are

networked into a small TCP/IP LAN, and have libraries and programs installed which allow processing to be shared among them.

There is no particular piece of software that defines a cluster as a Beowulf. Commonly used parallel processing libraries include MPI (Message Passing Interface) and PVM (Parallel Virtual Machine). Both of these permit the programmer to divide a task among a group of

networked computers, and recollect the results of processing. - Wikipedia

From PlayStation to Supercomputer for $50,000New York Times 2003-05-26 | By JOHN MARKOFF

As perhaps the clearest evidence yet of the computing power of sophisticated but inexpensive video-game consoles, the National Center for Supercomputing Applications at the University of Illinois at Urbana-Champaign has assembled a supercomputer from an army of Sony PlayStation 2's. The resulting system, with components purchased at retail prices, cost a little more than $50,000. The center's researchers believe the system may be capable of a half trillion operations a second, well within the definition of supercomputer, although it may not rank among the world's 500 fastest supercomputers.

Interconnect—Lowest measured latency (smaller better)

PathScale InfiniPath— 1.31 microseconds Cray RapidArray— 1.63 microseconds Quadrics— 4.89 microseconds NUMAlink— 5.79 microseconds Myrinet— 19.00 microseconds Gigabit Ethernet— 42.23 microseconds Fast Ethernet— 603.15 microseconds

Source: HPC Challenge, November 2005.

Latency and Bandwidth

Commodity processsors → the interconnect “makes” the supercomputer

TOP 10 Sites for June 2006

Site System Family # ProcessorsDOE/NNSA/LLNLUnited States

Blue Gene L

IBM

131 072

IBM Watson

United States

Blue Gene L

IBM

40 960

DOE/NNSA/LLNL

United States

ASCPurple (p –series)

IBM

12 208

NASA/Ames

United States

Altix

SGI

10 160

CEA

France

Tera-10 (SMP cluster)

Bull

8 704

Sandia Nat. Lab.

United States

PowerEdge

Dell

9 024

Tokyo Inst. Tech.

Japan

Sun Fire

NEC/Sun

10 368

FZ Juelich

Germany

Blue Gene L

IBM

16 384

Sandia Nat. Lab.

United States

XT3

Cray

10 880

Earth Simulation

Japan

NEC Vector

NEC

5 120

www.top500.org

Scientific Computing

Newton’s equation (with damping)

Diffusion or heat equation

Fourier transform

Schrödinger equation

Computational Needs- Biology & Bioinformatics

Problem Component Computing Speed Storage

Genome Assembly>10 TeraFlops sustained to

keep up with expected sequencing rates

300 TB of trace files per genome

Protein Structure Prediction >100 TeraFlops per protein set in one microbial genome Petabytes

Classical Molecular Dynamics 100 TeraFlops per DNA-protein interaction 10s of Petabytes

First Principles Molecular Dynamics

1 PetaFlops per reaction in enzyme active site 100s of Petabytes

Simulations of Biological Networks

>1 TeraFlops for simple correlation analyses of

small biological networks 1000s of Petabytes

Grand Challenges

Grand Challenges are the leading problems in science and engineering that can be solved only with the help of the

fastest, most powerful computers.

They address issues of great societal impact, such as

biomedicine, the environment,

economic competitiveness, and military.

Examples of Grand Challenge problems

Forecasting weather, predicting global climate change, and modeling pollution dispersion

Determining molecular, atomic, and nuclear structures

Understanding the structure of biological macromolecules

Improving communication networks for research and education

Developing and understanding the nature of new materials

Building more energy-efficient cars and airplanes

Understanding how galaxies are formed

Designing new pharmaceuticals

Blood flow in heart from Navier-Stokes equation, NIH

Brain Chemistry: bilayer sandwich of lipids — nitrogen (blue) and phosphorous (gold) heads facing outward on both sides of filamentary tails (gray). Patti Reggio, University of North Carolina, GreensboroDiane Lynch, Kennesaw State University

This thin-slice snapshot through the simulation volume, about 3 million light years thick by 4.5 billion light years on each side, shows the filamental structure of dark-matter clusters. Brightness corresponds to density.

Paul Bode and Jeremiah Ostriker of Princeton University

The infant universe hatching from its structureless shell. This map represents Edmund Bertschinger's simulations on the CRAY T3D at Pittsburgh. This map shows negative (blue) and positive (red) fluctuations of 0.0002 degrees K. The simulation assumed a mixed hot and cold dark matter model with 5 eV neutrino mass.

Von Mises stress on the F-16 structure (increasing from dark blue to light blue to red) during an aeroelastic simulation.

Mach contours and streamlines for an aeroelastic simulation at Mach 0.9.

Kelvin K. Droegemeier, University of Oklahoma at Norman.

Animation of a simulated earthquake in the San Fernando valley. Color depicts the peak magnitude of ground displacement. The

simulation covered a 54 x 33 kilometer area, superimposed here on a satellite view of topography, to a depth of 15 km. Los Angeles

can be seen in the southeast corner.

Carnegie Mellon University

Oxygen and Temperature in Turbine-Combustion

These snapshots from simulation of turbine-combustion show how oxygen decreases (previous) and temperature increases (this slide) as markers of combustion in turbineflow. DOE and Westinghouse

Adsorption of molecules at SAMs by theoretical/simulation modelling

Matrix Diagonalization

VectorScreening

Matrix Diagonalization

VectorScreening

Matrix Diagonalization

VectorScreening

Vector merge

MatrixGeneration

MatrixGeneration

MatrixGeneration

Convergence?

Stop

Vector Expansion

Vector Expansion

Vector Expansion

Initial Guess

no

yes

Parallel computing allows you todo calculations in new ways …

Conclusions …

• Lotsa uses for supercomputers

• Commodity processors, but a lot of ‘em

• Very fast customised interconnects

• Single winning architecture not decided

• Power remains a big concern- back to plumbing