Re: Running mpiBLAST on Microsoft Windows Compute Cluster Server 2003 with MS MPI

Hi Ray,
While I can comment directly on the Windows Compute Cluster OS, I 
personally have run mpiBLAST on a heterogenous windows 2000/XP/2003 
domain.  The biggest problem I encountered in that environment is that 
windows inexplicably kills all mpich jobs whenever a user logs on or off 
of one the compute nodes (which are actually just a set of workstations 
at my former lab).  Since we added the ability to resume a crashed or 
killed mpiblast run, we workaround the problem by running the same 
mpiblast job repeatedly until it has finished.  I have never used MS MPI 
or the compute cluster OS, so can't comment on whether the same problem 
crops up.  I'm sure there must be some windows API call that would 
prevent this nasty behavior, but googling and searching the API docs for 
a few hours didn't turn anything up before I lost patience. 

The windows mpiblast 1.4.0 is linked against mpich2, so one would have 
to recompile with the MS MPI libraries in order to use it instead.

Anybody else ever try mpiblast on windows?

-Aaron

Ray Muirhead wrote:
> Hi there,
>
> May I ask if anyone has experience with running mpiBLAST on Microsoft
> Windows Compute Cluster Server 2003, either with the MS MPI
> implementation or MPICH2?  Are there any potential problems that might
> occur with such a configuration?
>
> Thank you for your time.