HPCC: Information about Cray/SGI system.

The system consists of:

• newton.unicc.chalmers.se: SGI Origin 200 frontend machine in the following condiguration:
  • Four 180MHz MIPS R10000 CPUs with 1MB L2 cache rated at 10.3 SPECint95 and 15.6 SPECfp95 each;
  • 1GB RAM;
  • 9GB scratch disk;

• darwin(*): Two SGI Origin 2000 "backend" servers in the following configurations:
  Sixty four 195MHz MIPS R10000 CPUs with 4MB L2 cache rated at 11.4 SPECint95 and 19.1 SPECfp95 each; 15.5GB RAM; 25GB scratch disk (more to come); Forty 250MHz MIPS R10000 CPUs with 4MB L2 cache rated at 14.7 SPECint95 and 24.5 SPECfp95 each; 20GB RAM; 70GB scratch disk;
  (*) darwin is a common codename, Internet hostnames are unicorn.unicc.chalmers.se and popcorn.unicc.chalmers.se if you have to know.

The Origin 2000 system is fairly well suited for both large-scale parallel and serial jobs. The machine is binary compatible with the Challenge line of parallel servers from SGI, so existing applications tuned for this platform should run without modification. Users who want to develop their own parallel programs can take advantage of automatic loop parallelization carried out by the available compilers, or use message passing libraries such as MPI or PVM. POSIX-threaded applications are also supported.

• Latest news: availability reports, etc.

• Getting started

• Installed software

• User support

• Courses of interest for HPCC users

If you want to read more about the new system, a lot of documents written by SGI are available. For those who are looking for a brief introduction, the collection of on-line tutorials should be of interest.

In order to schedule and distribute batch jobs on the parallel system efficiently a new queue system is probably needed. For those who are interested, a summary of available scheduling systems is available here.

For those who are interested, we now provide the monthly generated statistics for all users on the web.

Lennart Bengtsson/Andy Polyakov/Paul Waserbrot