Header image  
By: Keishi Takashima  
   
   
 
The Application:

Super High Resolution Global Atmospheric Simulation Model (AFES) was ran on Earth Simulator for applications such as high-impact weather systems in East Asia, synoptic
disturbances in winter, the Baiu/Meiyu frontal.

how it uses parallelism
The AFES model utilized a three level parallelism method.
1) MPI-based parallelism is adopted for inter-node operations
2) Micro tasking for intra-node operations
3) Vector processing for a single processor
Your frank assessment of its successes and weaknesses.

Not just in Asia, but anywhere in the world there are natural disasters that have huge a huge impact on the socio-economic status of a country. For example, Hurricane Katrina was a disaster that took many lives and left the state in a horrifying scene. No one can stop a hurricane, but someone can anticipate it. If predictions on these disasters can be made, countries can prepare for the worst. Therefore using parallel computing technology to compute these predictions are in my opinion is the most important application in the current world.

Who did it?

Japan Agency for Marine-Earth Science and Technology

What kind of parallel computer did they use?

Earth Simulation Center located in Yokohama, Japan .Considered the fastest SC in 2002-2004.

640 processor nodes, where each node is a SX-6 super computer

SX-6 was designed by NEC

  • Upto 8 vector processors each with 72 vector registers.
  • Shares 64GB of memory
  • 0.15μ Cu CMOS process as a single chip
  • Peak performance of single node is 64 GFLOPS
  • Also include a scalar processor
  • The OS used is SUPER-UX an Unix like OS

Therefore the 5120 Vector Processors are available.
 11.7GB/s is obtained in inter-node communications.
A whole building is dedicated for this huge machine in Yokohama, Japan.

What programming language, tools, and/or libraries did they use?

For this model, FORTRAN 90, micro-tasking and MPI were used to take full advantage of the ES machine. The code for AFES consisted of approximately 30,000 Lines.

How did they measure and describe the performance of their code? How well did it perform?

AFES was developed to utilize the ES (Earth Simulation) at full potential.  The model has achieved performance of 26.58 Tera-Flops, which is still 64.9% of Max performance of the machine.

How big was the largest problem they ran, and how many processors did it run on?

The AFES model was used on a run labeled T1279L96. This covers 10km mesh and used all the processors on the ES( 640 nodes, 5120 proc).

 

Citation:

http://www.nec.de/press-room/news/news-detail/date////the-worlds-fastest-supercomputer-system-for-resolving-global-environmental-problems-completed/index.html

http://www.es.jamstec.go.jp/esc/research/AtmOcn/afes/afes.jp.html


http://www.jamstec.go.jp/res/ress/eno/pdf/ATHORPEX_proc.pdf