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: Brice Goglin (Brice.Goglin_at_[hidden])
Date: 2014-03-25 03:04:10


Le 25/03/2014 07:51, Biddiscombe, John A. a écrit :
>
> I'm compiling hwloc using clang (bgclang++11 from ANL) to run on IO
> nodes af a BGQ. 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'm a little worried because the first time
> I compiled, I had problems where apps would report an error from HWLOC
> on start and tell me to set HWLOC_FORCE_BGQ=1. when I did set this env
> var, it would then report that "topology became empty" and the app
> would segfault due to the unexpected return from hwloc presumably.
>

Can you give a bit more details on what you did there? I'd like to check
if that case should be better supported or not.

> I wiped everything and recompiled (not sure what I did differently),
> and now it behaves more sensibly, but with 15 instead of 16 sockets.
>
>
>
> Should IO be worried?
>

The topology detection is hardwired so you shouldn't worried on the
hardware side.
The problem could be related to how you reserved resources before
running lstopo.
Does lstopo --whole-system see more sockets?
Does BG_THREADMODEL=2 help?

Brice