[Bioclusters] weird MPI problem

Andrew Shewmaker bioclusters@bioinformatics.org
Thu, 22 May 2003 15:00:09 -0600


On Thursday 22 May 2003 02:43 pm, Jeremy Mann wrote:
> Maybe somebody on this list can help me figure a weird MPI problem. As
> most of you know, we run mpiblast, which has been working fine until this
> afternoon. We ran several blast jobs this morning, went to lunch, came
> back and started to do more blast jobs. Well, mpiblast is now erroring out
> and frankly I don't know why since it was working perfectly maybe an hour
> before. Anyway, here is the error I get:

Sometimes mpi can leave behind shared memory segments or semaphores.  
ipcs and iprm will show the status and remove ipc objects, respectively.
There are various ipc scripts out there that help you clean up lots of 
objects.

Some I haven't tried are at http://herdtools.sourceforge.net

cl_ipcs -- check on the Unix IPC information on remote machines 
cl_ipcrm -- remove Unix IPC segments from remote machines 

Hope this helps,

Andrew

-- 
Andrew Shewmaker
Associate Engineer, INEEL
Phone: 1-208-526-1415