Thoughts Beyond High Performance Computing: A Personal Assessment

HP-CAST 15 12 November 2010

Thoughts Beyond High Performance

Computing

A Personal Assessment

Dr Marek T. Michalewicz

A*STAR Computational Resource Centre

Singapore

(SC10 booth #4121)

Data Centre – HPC

ACRC Datacentre 1

Level 17 at Fusionopolis

Architect: Dr Kisho Kurokawa

ACRC at A*STAR

ACRC Datacentre 2

Matrix Building at Biopolis

A*STAR

Cores 7500

TFLOPS 75

Storage- HPC (attached to computing systems)- All research work

90 TB2 PB

System Loads- Ave- Low

95%75%

User Base 700

Projected growth rate 45%

# Data Centre (incl DR) 2 + 1 DR

New centres planned? Yes

# Staff 23

Note: Does not include private resources from RI’s

HPC at A*STAR at a glance

HPC at A*STAR: ACRC

• There are O(1) PFLOP systems, but O(1,000) TFLOP systems

CSIRO Advanced Scientific Computing Growth trends in HPC

CSIRO central computing systems

During the time I (MTM) worked there (ʼ90-ʼ00):

5 orders of magnitude increase in storage and > 2 in peak speed!

> 10 orders of magnitude increase in computational power in a life-span of one generation!

Slide courtesy of Dr R. Bell, CSIRO

ACRC at A*STAR

What’s the best method to accelerate your code?

ACRC at A*STAR

Do nothing - just wait for faster computer ....

What’s the best method to accelerate your code?

ACRC at A*STAR

Do nothing - just wait for faster computer ....

12 years later: No of atoms ~ 8x10^9

ACRC at A*STAR

Do nothing - just wait for faster computer ....

12 years later: No of atoms ~ 8x10^9

machine X (estimated) 2010 ~7,500,000,000 1,953,125 ~17,000 172,000

ACRC at A*STAR

20

40

60

80

100

2002 2003 2004 2005 2006 2007 2008 2009 2010

ACRC+ Fuji+ SMPIHPCBIIextras

A*STAR HPC computational power

HPC Resources at A*STAR: ACRC

TFlops

ACRC at A*STAR

400

800

1200

1600

2000

2002 2003 2004 2005 2006 2007 2008 2009 2010

IHPCBIIACRC

A*STAR HPC data storage

HPC Resources at A*STAR: ACRC

TBytes

ACRC at A*STAR

1,000,000

2,000,000

3,000,000

4,000,000

5,000,000

2002 2003 2004 2005 2006 2007 2008 2009 2010

IHPCBIITOTAL/ACRC

A*STAR HPC CPU Utilisation (CPU hours/month)

HP-CAST 15 12 November 2010

A computer is a programmable machine that receives input, stores and manipulates data, and provides output in a useful format. (Wikipedia)

HP-CAST 15 12 November 2010

Past

HP-CAST 15 12 November 2010

HP-CAST 15 12 November 2010

Antikythera 150–100 BCThe Antikythera mechanism, is conjectured to be an ancient mechanical computer designed to calculate astronomical positions. It was recovered in 1900–01 from the Antikythera wreck, but its complexity and significance were not understood until decades later.

animations at:http://www.mogi-vice.com/Antikythera/Antikythera-en.html

HP-CAST 15 12 November 2010

Jacquard loom: Joseph Marie Jacquard 1801

Importance to Computing

The Jacquard loom was the first machine to use punched cards to control a sequence of operations. Although it did no computation based on them, it is considered an important step in the history of computing hardware. The ability to change the pattern of the loom's weave by simply changing cards was an important conceptual precursor to the development of computer programming. Specifically, Charles Babbage planned to use cards to store programs in his Analytical engine.

HP-CAST 15 12 November 2010

Thomas Fowler 1840In 1840 Fowler produced a mechanical calculating machine which operated using ternary arithmetic. He designed and built the machine single-handed from wood in the workshop attached to his printing business. To compensate for the limited precision achievable using wooden components, he constructed the machine on a large scale; it was 6 feet long by 3 feet deep and 1 foot high (1800 x 900 x 300 mm).

HP-CAST 15 12 November 2010

Charles Babbage FRS (1791 –1871)Analytical engine 1834-1871

Difference engine

The Difference Engine was an automatic, mechanical calculator designed to tabulate polynomial functions. Both logarithmic and trigonometric functions can be approximated by polynomials, so a difference engine can compute many useful sets of numbers.

As soon as an Analytical Engine exists, it will necessarily guide the future course of the science.—Passages from the Life of a Philosopher, Charles Babbage

HP-CAST 15 12 November 2010

This Difference Engine is not a replica, there never was one built during Babbage's lifetime. This is the first one, the original built over 100 years after Ch. Babbage death!

HP-CAST 15 12 November 2010

A selection of seven Enigma machines and paraphernalia exhibited at the USA's National Cryptologic Museum. From left to right, the models are:

1) Commercial Enigma; 2) Enigma T; 3) Enigma G; 4) Unidentified; 5) Luftwaffe (Air Force) Enigma; 6) Heer (Army) Enigma; 7) Kriegsmarine (Naval) Enigma—M4.

Enigma machine

HP-CAST 15 12 November 2010

The bomba, or bomba kryptologiczna was a special-purpose machine designed about October 1938 by Polish Cipher Bureau cryptologist Marian Rejewski to break German Enigma-machine ciphers.

Other contributors: Henryk Zygalski, Jerzy Różycki

In December 1932, Marian Rejewski made what historian David Kahn describes as one of the greatest advances in cryptologic history, by applying pure mathematics – group theory – to breaking the German armed forces' Enigma machine ciphers.

HP-CAST 15 12 November 2010

Howard Aiken Harvard Mark 1 1944

HP-CAST 15 12 November 2010

His greatest achievement was the world's first functional program-controlled Turing-complete computer, the Z3, in 1941 (the program was stored on a punched tape).

Konrad Zuse 1910 – 1995

Konrad Zuse's electromechanical "Z machines". The Z3 (1941) was the first working machine featuring binary arithmetic, including floating point arithmetic and a measure of programmability.

In 1998 the Z3 was proved to be Turing complete, therefore being the world's first operational computer.

HP-CAST 15 12 November 2010

Fujitsu Facom 128Japan's first relay-based commercial

computer – 1956

HP-CAST 15 12 November 2010

Present

HP-CAST 15 12 November 2010

Cloud computingGrid computingGoogle ?Internet as a computer special purpose architectures

grape 3, anton, ..........teramac (HP project - culled?)Cell Matrix

accelerators FPGAGPU

HP-CAST 15 12 November 2010

Connects

Direct Broadcast Optical Interconnect (DBOI)simultaneous, all-to-all, optical interconnect from Lightfleet

http://lightfleet.com/

HP Research LabsKuekes, et al, Stan Williams

HP-CAST 15 12 November 2010

Future

HP-CAST 15 12 November 2010

21st Centure computers:non-von Neumann digital - non-binaryanalogmixedcellular wave computerscellular nonlinear networks (CNN)molecular computerquantum computer

Exa FLOPS computer(and 1000s Peta FLOPS computers)

but - what embodiment, architecture?

HP-CAST 15 12 November 2010

computers +

sensors (networks)

HP-CAST 15 12 November 2010

Sonobuoys and sonars: Navy, environment and biology

Displacement and tremor sensors in security perimeter systems, border protection, cargo monitoring, seismology, mining, geology, tectonics and nuclear test monitoring

Vibration meters attached to disk drives in datacentres - used to detect earth quakes

Accelerometers: manufacturing, aviation, defense, aeronautics and automotive

Sensors - some applications

HP-CAST 15 12 November 2010

Brave new world of (nano)-sensing

• Sensing methods to effectively help de-mine unexploded land-mines and shells

(Egypt, Iraq, Afghanistan, ....)

• “ultra-sound” scans of Pyramids and other structures

• 24/7 real-time monitoring of health of water dams, bridges, roads, railway tracks

• Continuos sensing and control of environmental conditions “at very small

granularity” - localised scale

• Point-of-delivery (plant) moisture sensing and watering systems

• Bio-medical diagnostics applications

• Environmental sensors embedded in mobile phones

• Oil & Gas: seismic, reservoir, well, “Smart” Oil Fields

• Tsunami, earthquake early warning

• etc...

HP-CAST 15 12 November 2010

University of California, Berkeley “Smart Dust project”

HP’s CeNSE project: “Create the mathematical and physical foundations for the technologies that will form a new information ecosystem, the Central Nervous System for the Earth (CeNSE), consisting of a trillion nanoscale sensors and actuators embedded in the environment and connected via an array of networks with computing systems, software and services to exchange their information among analysis engines, storage systems and end users.”

Foresight Institute: Open Source Sensing Initiative

The University of Washington Pacific Ocean floor remote sensing using optical fibre cables and swarms of autonomous vessels laced with sensors and observation devices.

IBM, Fujutsu, HP, ..... all talk and work towards sensing the world.

More examples of the trend:

HP-CAST 15 12 November 2010

The greatest obstacle

In-silico

Prototyping

Fab costs

Time of development of commercially viable processors -

not to mention different architectures or concepts ....

Software creation

HP-CAST 15 12 November 2010

Summary:

Future of sensor networks:

In few years from now sensor networks will be as ubiquitous and pervasive as cellular phones are today.

They will require massive amount of computing power.

HPC systems:

Before Exascale system is build there will be 1000 Petascale systems.

Architecture:

Our extrapolations based on current state-of-the-art solutions might be quite misplaced.