Open MPI logo

Hardware Locality Development Mailing List Archives

  |   Home   |   Support   |   FAQ   |  

This web mail archive is frozen.

This page is part of a frozen web archive of this mailing list.

You can still navigate around this archive, but know that no new mails have been added to it since July of 2016.

Click here to be taken to the new web archives of this list; it includes all the mails that are in this frozen archive plus all new mails that have been sent to the list since it was migrated to the new archives.

Subject: Re: [hwloc-devel] [hwloc] #12: support user-defined processor restriction
From: Ashley Pittman (ashley_at_[hidden])
Date: 2010-02-15 17:16:16


On 15 Feb 2010, at 22:02, Samuel Thibault wrote:

> Ashley Pittman, le Mon 15 Feb 2010 21:52:29 +0000, a écrit :
>> On 15 Feb 2010, at 21:46, Samuel Thibault wrote:
>>>> I say the commit r1726 which closed #21 and am working on testing this now, it certainly appears to be what I requested.
>>>
>>> Maybe, depending on whether you want to discover according to the other
>>> process' binding (sched_setaffinity) or according to the other process'
>>> restricted view of the machine (Linux cpuset).
>>
>> I don't understand the difference, I thought they were two ways of achieving the same thing?
>
> Linux cpusets are administrator restrictions. A normal user can not
> escape its cpuset. It can however choose the binding within that
> cpuset. That is why by default hwloc limits its topology detection
> to the cpuset of the current process, since it's not allowed to bind
> threads of the process beyond that anyway. (same for the set of online
> cpus).

Ah that makes sense.

I suspect the right thing to do is to from my perspective is to report the affinity rather than the cpuset then or rather whichever of the two is most restrictive which will be the affinity if it's set and the cpuset if it's not.

Ashley,

-- 
Ashley Pittman, Bath, UK.
Padb - A parallel job inspection tool for cluster computing
http://padb.pittman.org.uk