Open MPI logo

Hardware Locality Users' Mailing List Archives

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

Subject: Re: [hwloc-users] [EXTERNAL] Re: hwloc_get_latency() failures and confusion
From: Wheeler, Kyle Bruce (kbwheel_at_[hidden])
Date: 2012-08-06 18:36:58


On Aug 6, 2012, at 4:08 PM, Brice Goglin wrote:

> You should not be getting a distance matrix for depth 0 above. You get
> one for depth=1 (the depth of NUMAnodes in your topology).

Ahh, ok. Well, at least I know what I *should* get now. :)

Sheepishly, I have to admit that I just figured it out. I had two versions of hwloc installed, and the older one was further up in the LD_LIBRARY_PATH.

>> And why is hwloc_get_latency() failing?
>
> If you pass some Core objects to get_latency(), it's expected that it
> fails because the topology only has latencies between NUMA nodes. You
> should walk up the object parent links until you find NUMAnode objects.
> We've been thinking of handling this case inside hwloc but we're not
> sure it's always a good idea to do so.

Ahhhh, that's key! The documentation currently says "Look at ancestor objects from the bottom to the top until one of them contains a distance matrix that matches the objects exactly", which suggests to me that it will traverse the object hierarchy looking for the NUMANodes *for* me. Now that I know to do it manually, I can get it to work. (Of course, my version mis-match didn't help in the first place…)

Thanks for the help!

-- 
Kyle B. Wheeler
Dept. 1423: Scalable System Software
Sandia National Laboratories
505-844-0394