When ever I try to look at my project at the minute I feel like I'm striking my head against a brick wall. I managed to find one Math paper suggesting that the Conjugate Gradient method can be parallelised onto distribuded memory systems, but when I compared the data they were feeding it, to the data I'm using I don't think it's relevant, my data is just too damned sparse. Which means that the time I spend sending any results around totally outweighs the time it takes to just do the extra calculations. I think that my code might possibly be of some use. But only if you're using a REDICULOUSLY large data set. My one chance so far to work on a less sparse system was with a 3D model rather than the 2D ones I've been using. Unfortunatly the matrix I was given is not suitable for the CG method. Argh! And to top it all off my project supervisors not in Bath at the minute. Greh.
The only thing I can think of left to do on this project is possibly to try and use a dual CPU machine and share the memory, to see how well things would scale that way... (Any one have a multi-CPU machine I could aquire some CPU time on?) I suppose that another option ma be to create some "fake" test data to feed into the system to see quite where the usefulness of my project might lie, but I have no idea how to go about generating sparse symetric matricies.