On Wed, 2007-07-11 at 19:45 +0200, Bert Wesarg wrote:
>
> Li-Ta Lo wrote:
> > On Wed, 2007-07-11 at 13:29 -0400, Jeff Squyres wrote:
> >> On Jul 11, 2007, at 1:24 PM, Li-Ta Lo wrote:
> >>
> >>> On Wed, 2007-07-11 at 12:40 -0400, Jeff Squyres wrote:
> >>>> Hmm. So I'm confused -- I thought you said in #9 that "There should
> >>>> be just 2 physical sockets on the mainboard". So PLPA is now
> >>>> reporting 2 sockets, numbered correctly.
> >>> I didn't say how plpa-info should report. I was giving you the output
> >>> of the plpa-info and it was wrong w.r.t what you are trying to fix. It
> >>> says there are 3 sockets which is still wrong.
> >> Hah! I missed that; my bad.
> >>
> >>>> [ollie_at_exponential plpa-info]$ ./plpa-info --topo
> >>>>> Kernel affinity support: yes
> >>>>> Kernel topology support: yes
> >>>>> Number of processor sockets: 3
> >>>> I do have PLPA intentionally skipping reporting sockets that have no
> >>>> cores on them. Are you saying that plpa-info should report sockets
> >>>> with no cores? I chose not to do that because how would you know
> >>>> what they are? Linux only reports the sockets that are in use -- how
> >>>> would you know what sockets are not in use? For example, what if you
> >>>> populate a motherboard with 4 sockets in sockets 0 and 1 -- how would
> >>>> PLPA know that sockets 2 and 3 exist?
> >>> The output still talks about socket without cores. It is right with
> >>> socket 1 (not counting/reporting it) but it is wrong with socket 2.
> >> Is the tarball attached to #9 still accurate? I was using that when
> >> testing the new stuff and I was getting 2 sockets displayed from plpa-
> >> info, and only showing 1 core on sockets 0 and 3.
> >>
> >
> > No. I turned of SMT yesterday when answering you bug report.
> > How should I create the tarball again? Just tar czf ?
> This was my howto:
>
> http://www.open-mpi.org/community/lists/users/2007/04/3126.php
>
Here it is.
Ollie
|
![[Flag]](../../../../../community/mirrors/flag-us.gif){kind=small}