?

Log in

No account? Create an account
Project

Project update.

In an attempt to aleviate the boredom of mstevens. Here's a post on how my project appears to be going.

Well I now have code that works on the following setups; A Beowulf cluster with single processor machines, talking over MPI; An SMP Linux box splitting the load processors, using pthreads; and code that appears to work on a uniform system of SMP machines clustered together with MPI.

Gah some of the bugs were really dumb mistakes.

Comments

Do you have MPI coexisting with pthreads, then? Was it painful? Do you have a high-level API so that one can write code that doesn't paricularly care how it's being parallelised?
Yeah, MPI is quite happily co-existing with pthreads. MPI doesn't care what the program that's running and communicating is. All I've done is run a pthreaded program using the MPI architecture to communicate between two programs and split the workload across the machines, then the workload that's been split by machine is further split into threads.

It wasn't that painful, any more than using either one seperately, with annoying bugs that show up in the strangest ways...

I'm using the MPIch library, and the stanard Linux POSIX pthreads API. With MPI all I tell it is the data I wish to either broadcast to all nodes, or to specific nodes, so I wouldn't call it all that low level...
fireworks

August 2010

S M T W T F S
1234567
891011121314
15161718192021
22232425262728
293031    
Powered by LiveJournal.com