[mpich-discuss] Binding ranks to CPUs + Hydra-Bug

Ken Raffenetti raffenet at mcs.anl.gov
Wed Dec 18 13:15:57 CST 2013


I see. I confirmed this on my laptop. That is indeed a bug we'll need to address.

Ken

----- Original Message -----
> From: "Jan Bierbaum" <jan.bierbaum at tudos.org>
> To: discuss at mpich.org
> Sent: Wednesday, December 18, 2013 1:02:04 PM
> Subject: Re: [mpich-discuss] Binding ranks to CPUs + Hydra-Bug
> 
> On 18.12.2013 19:49, Ken Raffenetti wrote:
> > You will have to have bindings specified for all your ranks in the
> > case you describe.
> I would like to, but whenever the list of comma-separated CPU ids
> contains more values than there are CPUs in the system I get an
> assertion failure.
> 
> For example:
> My test system has 4 CPUs. If I try to run 'mpiexec -np 9 -bind-to
> user:0,1,2,0,1,2,0,1,2 /some_path/some_binary' Hydra will crash and
> show
> the error I pasted in my original message.
> 
> 
> Regards, Jan
> _______________________________________________
> discuss mailing list     discuss at mpich.org
> To manage subscription options or unsubscribe:
> https://lists.mpich.org/mailman/listinfo/discuss
> 



More information about the discuss mailing list