[mpich-discuss] Multiple MPI_Comm_spawn to a single multicore host

[Roberto Fichera]

Rajeev Thakur ha scritto:
> It is supposed to work even with multiple processes per node. Are you
> creating the multiple processes with a single MPI_Comm_spawn (and maxprocs
> set appropriately) or with multiple calls to MPI_Comm_spawn. It should work
> either way.
>   
I use only one MPI_Comm_spawn() per slave for spawning *only* one slave 
process. If the target node
is the master, I see the problem.
> Rajeev 
>
>   
>> -----Original Message-----
>> From: owner-mpich-discuss at mcs.anl.gov 
>> [mailto:owner-mpich-discuss at mcs.anl.gov] On Behalf Of Roberto Fichera
>> Sent: Tuesday, July 15, 2008 10:29 AM
>> To: mpich-discuss at mcs.anl.gov
>> Subject: [mpich-discuss] Multiple MPI_Comm_spawn to a single 
>> multicore host
>>
>> Hi All on the list,
>>
>> I'm using the MPI_Comm_spawn() to dynamically distribute the 
>> jobs in our 
>> nodes. Everything
>> works fine until I use one job per node even if the target node is a 
>> multicore/smp machine.
>> Whenever I start to spawn multiple jobs, as the number of cpus on the 
>> target machine, than I
>> start to see crashes on the MPI application, maybe due to the mpd 
>> shutdown caused by the first
>> MPI_Finalize() issued by the slave. So, does anyone know how 
>> to handle 
>> correctly such distribution.
>> I'm currently using MPICH2 v1.0.7 within a Fedora 8 x86_64 OS.
>>
>> Thanks in advance.
>> Roberto Fichera.
>>
>>
>>     
>
>
>   

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/mpich-discuss/attachments/20080718/034c53bb/attachment.htm>