Linux hackers first put together a cluster of PlayStation systems a few years ago, but it has not gained much attention. It is an interesting application of a game console and is certainly viable for some applications (such as Computational Chemistry in this case), but it is not likely to find much use for running CFD codes due to the small amount of memory available.

There are generally two reasons to run in parallel:

1) Increased problem size: maxing out the memory on multiple systems allows you to run significantly larger problems;

2) Increased speed: by spreading the workload you can get to a solution faster.

The PlayStation2 has only 32 MB of RAM (see http://www.psx2central.com/misc/psx2info.htm) and is not expandable. This would mean that you would require at least 64 PlayStations to make up the RAM available on a typical PC, not very economical.

As for parallel speadup, the parallel efficiency of CFD is very problem size dependant. A typicall CFD code solves multiple "chunks" or partitions of the mesh and relies on inter-CPU communication at the boundaries between partitions. If you are limited to 32 MB of RAM, and considering that some of this is taken up by the executable itself, the CPU will be sitting idle most of the time waiting for data to pass through the network bottleneck.

So in response to the question "When will we see the first PS/2 cluster running a CFD code?", I can't say that nobody would try it, but I find it doubtful that we will see any useful application of PlayStation 2 clusters to CFD.

Robin

• Sony PS/2 Clustering - News Editor, Tue, 30 Jul 2002, 11:15 a.m.
  • Re: Sony PS/2 Clustering - Robin, Tue, 6 Aug 2002, 3:56 p.m.