Open MPI logo

Open MPI User's Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Open MPI User's mailing list

From: Bo Peng (ben.bob_at_[hidden])
Date: 2007-02-05 12:22:39

On 2/5/07, Tom Rosmond <rosmond_at_[hidden]> wrote:
> Have you looked at the self-scheduling algorithm described in "USING
> MPI" by Gropp, Lusk, and Skjellum.

Yes, a master-slave mode should be better for my module. In this way:

1. the master is started on node 0 and execute the Python script, the
master send command to the slaves if it needs to manipulate data on a
particular node.
2. the slave turns to a 'feed-me' loop after the module is loaded. It
accepts something like an operation code and parameters, then look up
a function and execute.

My biggest problem now is that the types of operations are numerous
and the programming mode would be vastly different between master and
slaves. I mean, the module is written in C++ and is wrapped by SWIG so
that I have complete C++ class hierarchy. How much the slaves should
*undertand* the data structure is a problem.

Anyway, many thanks for your suggestion.