[mvapich-discuss] Dual FDR setup launches multiple instances of same process

Hari Subramoni subramoni.1 at osu.edu
Mon Mar 3 09:43:02 EST 2014


Hello Santak,

The job launch process is not influenced by the number of HCA's in the
system. The job launcher (I believe it is Hydra in this case) should not be
launching more processes than what is requested by the user through the
"-n" command line option. We ran the exact command you gave and have
verified that it works as expected on our local systems with one and two
HCAs.

Could it be that there are some stray processes from previous runs or from
other users trying to use the system concurrently? Could you also let us
know which version of MVAPICH2 you are using and how you configured it?

Regards,
Hari.


On Mon, Mar 3, 2014 at 1:09 AM, Dalai, Santak
<Santak.Dalai at kla-tencor.com>wrote:

>  Hello,
>
>
>
> I have a setup where I have two nodes, each having two FDRs. All FDRs are
> connected to an FDR switch. I am trying to run simple programs like
> osu_put_bibw. I observed that it is creating three instances of the same
> process. Has anyone observed similar behavior?
>
> I wanted to understand why it's launching 3 instances. Can someone explain
> how a process is launched when we run using multiple HCAs?
>
>
>
> I am using 'mpirun -n 2 -f mf1 -env MV2_NUM_HCAS=2
> ~/mvapich2-2.0b/libexec/mvapich2/osu_put_bibw' command to launch the
> process.
>
>
>
> Here is the snapshot of the top command
>
>
>
> If we notice one of one if the processes utilizes more CPU than others.
>
>
>
> Thanks,
>
> Santak Dalai
>
>
>
> _______________________________________________
> mvapich-discuss mailing list
> mvapich-discuss at cse.ohio-state.edu
> http://mailman.cse.ohio-state.edu/mailman/listinfo/mvapich-discuss
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cse.ohio-state.edu/pipermail/mvapich-discuss/attachments/20140303/1782ed9d/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image002.png
Type: image/png
Size: 24249 bytes
Desc: not available
URL: <http://mailman.cse.ohio-state.edu/pipermail/mvapich-discuss/attachments/20140303/1782ed9d/attachment-0001.png>


More information about the mvapich-discuss mailing list