How to create a distributed GPU network like bitcoin for number crunching?

[marvinmartian]

I have a bunch of algorithms I'd like to run, testing various parameters, using something like the distributed model of bitcoins (without hashing or security) so that when I'm not mining, I can use the HP of my rig to crunch numbers.

Any good pointers on how to get started on something like this?  What I'm mainly looking for is two things:

1.  How to interface with OpenCL (which I can likely get from AMD).

2.  How to make it a distributed model where a central server says something like the following:

   return_data[tag] = run_on_network_cluster(my_function(), param1, param2, param3, param4, ...);

So that I can collate all the data in the end and sort it at the end.  EG., each run will have a unique output.

Ok, so just one thing.  ;-)

Thanks.

MM

[wasabi]

Write code.

[entens]

I'm going to let you sift out what you want, because the BOINC project has almost all the information you'll need: http://boinc.berkeley.edu/trac/wiki/SoftwareDevelopment.