itaps-parallel Notes from today's meeting

[Mark Beall]

On Apr 19, 2010, at 10:06 AM, tautges at mcs.anl.gov wrote:

>> Eventually, in the hyperparallel world that's coming, apps and  
>> implementations may both have to be written to be multithreaded  
>> shared-memory processes within a node (OpenMP?) with communication  
>> between nodes (MPI?).  This isn't in conflict with the notion of  
>> exactly one part (and iMesh instance) per process; it just requires  
>> more careful programming of apps/services that use the interface.
>
> Imeshp is written in terms of message passing anyway, so it'll need  
> to change to handle threads anyway.

It depends on how you want to approach using threads. Our  
implementation (of our partitioned mesh, mesh generation, mesh  
improvement and mesh adaptivity) use either MPI parallel or multiple  
threads and the abstraction is the same. There are some good and bad  
points to this (although nothing about it precludes using threads in a  
different way in addition). In theory, we could use this for MPI  
parallel with multiple threads per process, however we've never done  
that due to lack of a thread-safe MPI implementation (not sure if  
there is one these days - we could work around the lack of it with  
additional effort) and lack of a reason to do it.

mark