Yeah they've moved it to be fully compatible with everything up to and including Java, so there is lots of it around. I see it in modern clusters though there other effective approaches, what we're really suffering from is a lack of new methodology. This is changing though as multi core systems become a way of life. It will take time, but we'll see it happen. New tools are needed that's for certain!

I agree with you fully on why they made the choice for MPI, I think coding resources at Stanford are quite limited, they needed something they could leverage across platforms and this was it. MPI is HIGH on the list of compatibility items MS would love to drop, trust me, so I'm not sure how many optimizations we'll see for it.

LOL@ distributed memory, I know what you're saying and yeah, if you think back to the 70's when MPI was put into play ... all memory was distributed and non allocated to a processor. You had to code your app to use a quantity of that resource. Processor were virtual for the most part, just based on time slices etc.. an entirely different model but here is MPI still plugging away, pretty wild when you think of it.