I'm setting up a compute cluster that uses MPI to generate calculations on the fly in short amounts of time <3s. I was wondering what kind of overhead times I would possibly see if I submitted jobs using the CCP API, or should I just use the Cluster.ExecuteCommand method instead?
If neither of these is a possibility because the overhead of the CCP framework is too large, is there a way to bypass that and still harness the power of the MPI.NET?
You may want to investigate our SOA model. It may be better suited for your scenario, since it allows you to start up a service and then make multiple calls in to it without having to pay a start-up cost each time like you would if you were launching a new MPI app every couple seconds.
You'll find a bunch of resources on this on microsoft.com's technet & msdn libraries, here is one starting point http://www.microsoft.com/downloads/details.aspx?FamilyId=99742128-4C8C-4ADB-872E-C4099206C78D&displaylang=en