need programming help for cluster

[Mad_Scientist]

So I have a peice of software that runs a client on network computers. The software searches the network for these clients and then distrubutes its task out to each pc.

So this is a Bonjour application (client) that runs - as far as i know - only on windows... does anyone have ANY idea how hard it would be to overlay this client on the parallela? Maybe port it to linux and run it inside an actual OS.


There is a TON of people who would monetarily contribute to this project. for the price of the board, if we could get the software working I know my company would probably buy as many as i asked them to. id say 16 or so would be great and thats just one company.


so..can i get some direction on this?

[tincman]

Avahi is a Zeroconf (and Bonjour compatible I think?) implementation for Linux, so unless you have other Windows specific parts the porting should hopefully be fairly straigtforward.

The other option that might be quicker to deploy would be to run it under Wine (a "non-emulator" that translates Windows-API calls into compatible Linux ones). I imagine that Bonjour runs well enough under Wine and/or Wine translates Bonjour API calls into appropriate Avahi ones.

Now, to actually use the Epiphany chip would be the more useful part, and this depends on what kind of work the client does.

Best case scenario, does this client use OpenCL? In which case both options will work--the Epiphany chip should be exposed as an OpenCL device, and the stdcl library will compile your kernels to utilize it. If you link OpenCL in a standard way, the port should be straightforward. I'm not sure about how functional Wine's OpenCL support is, but I do know it exists and thus you should be able to run the client under wine and still have it accelerated.

Next best case scenario, you wait for the community to make and/or make yourself an Epiphany accelerated backend library (maybe a BLAS/LAPACK library? Depends on what the client does I suppose). This would require porting the client to Linux, but might make the porting easier as you just link in the new library/backend.

Worse case scenario, the computations/logic are very specific, in which case you would need to modify the algorithms to properly use the cores (however, the general nature of the cores makes this not too difficult, but it'd be close to programming for the Cell BE).

Hope this helps

[9600]

[tincman]

[Mad_Scientist]

I dont have access to the source, and to be honest i highly doubt i can get it. this is a rendering program in CAD that is based on the luxology engine. as far as i know its just crunching numbers and its all CPU core based.. (no gpu untilization at all sadly)

Somehow they have it locked down where the only way you can use network rendering is if you have a subscription. (which we do) but i dont think they would let the source go, just for that aspect alone..

Dang.. i really thought this would work..

[Gravis]