Open MPI logo

Hardware Locality Users' Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Hardware Locality Users mailing list

Subject: Re: [hwloc-users] BGQ question.
From: Jeff Hammond (jeff.science_at_[hidden])
Date: 2014-03-25 09:22:51


I am inclined to think the issue us that ion Linux uses only 16 cores. When hwloc does n-- for CNK to skip the 17th (OS) core, it gets the wrong answer for Linux.

Just check for Linux support and use /proc/cpuinfo and don't adjust manually.

I'm not sure hwloc on BGQ ion ion needs and special hook either. Isn't Linux PPC support enough?

Jeff

Sent from my iPhone

> On Mar 25, 2014, at 7:28 AM, Chris Samuel <samuel_at_[hidden]> wrote:
>
>> On Tue, 25 Mar 2014 06:51:49 AM Biddiscombe, John A. wrote:
>>
>> I’m compiling hwloc using clang (bgclang++11 from ANL) to run on IO nodes af
>> a BGQ.
>
> Out of interest, why on an I/O node?
>
>> It seems to have compiled ok, and when I run lstopo, I get an output
>> like this (below), which looks reasonable, but there are 15 sockets instead
>> of 16.
>
> I've not tried on our I/O nodes, but looking at /proc/cpuinfo on one it
> reports 68 cores (hardware threads), thus 17 real cores (on CNK one of those
> is reserved for the CNK so is not available for codes without determined
> fiddling).
>
> -bash-4.1# grep -w processor /proc/cpuinfo | wc -l
> 68
>
> This is V1R2M0 (RHEL 6.3 based).
>
> cheers,
> Chris
> --
> Christopher Samuel Senior Systems Administrator
> VLSCI - Victorian Life Sciences Computation Initiative
> Email: samuel_at_[hidden] Phone: +61 (0)3 903 55545
> http://www.vlsci.org.au/ http://twitter.com/vlsci
>
> _______________________________________________
> hwloc-users mailing list
> hwloc-users_at_[hidden]
> http://www.open-mpi.org/mailman/listinfo.cgi/hwloc-users