[mvapich-discuss] Configuration with Many code nodes

[Krishna Kandalla]

Hi Ralph,
            _AFFINITY and VIADEV_USE_AFFINITY are enabled by default in
MVAPICH.  However, if you have a scheduler that can assign process to core
affinity, it is possible that it can conflict with what we do within our
library. In such cases, you are probably better off disabling MPI-level
affinity in your launcher script and letting the scheduler handle it
completely. On the other hand, if the scheduler does not do it, you can
always use VIADEV_CPU_MAPPING to indicate a specific process to core binding
pattern. Please let us know if you have any other questions.

Thanks,
Krishna

On Tue, Aug 16, 2011 at 8:14 AM, Ralph C. Bording <rcbord at vims.edu> wrote:

>  Hi All,
>   I am trying to install the latest MVAPICH-1-2 on our new cluster. The new
> nodes have 48 cores and with these
> new nodes I can see the mpi processes hopping from core to core.  We are
> using Torque and Maui to enable cpusets but within the cpusets we need to be
> able to configure MVAPICH so that is "USE"s  AFFINITY.  From the
> documentation it looks like I want to set the parameter VIADEV_USE_AFFINITY
> to do this.
>
> Is this conceptually correct?  Do I need to do anything else?  Can I set
> VIADEV_USE_AFFINITY to default so that every job runs this way or will have
> have to have all the users modify their PBS scripts accordingly?
>
> Lastly in the documentation is has, "VIADEV_USE_AFFINITY does not take
> effect when _AFFINITY_ is not defined."    So my question is how do I define
> "_AFFINITY_" ?
>
> Thanks.
> RBording
> HPC Application Analyst
> Virginia Institute of Marine Science
>
> _______________________________________________
> mvapich-discuss mailing list
> mvapich-discuss at cse.ohio-state.edu
> http://mail.cse.ohio-state.edu/mailman/listinfo/mvapich-discuss
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.cse.ohio-state.edu/pipermail/mvapich-discuss/attachments/20110816/12d9a88e/attachment.html