The SMPs future

[Ahavi]

This is how I see the situation. All new PCs are shipped with dual/quad cores. The main load is handled by non computer enthusiasts (set it and forget it). The development for an easy to use, stable, set it and forget it SMP client is essential. The SMP client cannot handle the mainstream WUs regular clients can. How is this suppose to work out if you think the mainstream WUs are really important? I think it will be just like 5.04 where you can select how large the tasks should be (if you're a F@H/computer enthusiast running 24/7), plus an option where it automatically selcets based on neccesity (e.g 2x 5.04 WUs handled by one SMP). What do you guys think of the SMPs future, particulary in regards to WU sizes, points and deadlines?

[VijayPande]

SMP can handle the mainstream WU's, but don't share the same file format. As more people get SMP-capable boxes and the SMP code becomes more stable, we hope donors will migrate to the SMP client. The SMP client is pretty stable for OSX and Linux, but not as much for Windows (since Windows really wasn't meant to run supercomputer-class applications like MPI). However, the newest mods to the Windows SMP has been moving this in the right direction, and we're considering other alternatives all the time.

[toTOW]

We're considering other alternatives all the time.

Multithreads instead of multiprocesses ?

[Foxery]

You've smooshed two unrelated issues into one thought.

1) The SMP client isn't stable, and this limits who can run it because someone needs babysit the machine. This will be fixed in time with new versions of the program.

2) It receives different projects/WUs than the old client. This isn't a problem, per se. It will also become less true over time as old machines continue to be replaced and shut down. Stanford will release fewer new projects for the old clients to compute, and more for new clients. No work will be left unfinished.