views:

227

answers:

4

For relatively coarse-grained parallel processing (data transfers can be done in 100 Mbit Ethernet network), which middleware that is available for Windows and Linux do you consider best? CORBA? MPI1? MPI2? XML-RPC/SOA/WSDL? Custom network protocols? Java RPC? Python RPC? Others?

A: 

PVM was very useful to me for a small (10 machine, 40 processor) cluster with a GigE switch. This was a couple years ago, but should still be viable, though PVM does not seem to be evolving any further. I mention it to add it to you list of candidates. I suspect MPI would be a more popular choice, but I liked PVM.

Doug Currie
A: 

I dont understand why you compare MPI with CORBA or SOA with RPC ? Could you explain what you want ?

Well, MPI, CORBA, XML-RPC etc. are all middleware solutions that can be used for parallel processing. If you have fine-grained data parallelism you can't use e.g. XML-RPC because of its overhead, but as I mentioned in the question, that's not the case.
antti.huima
A: 

Or may be Terracotta? Or may be Hadoop?

Artyom Sokolov
+1  A: 

You're asking about middleware, but then specifying a series of messaging protocols. If you're looking for messaging protocols, the Thrift framework from Facebook is definitely something that works under very high load and cluster size:

Thrift

If you're looking for some other component of middleware, please clarify.

runako

related questions