[mpich-discuss] cpu binding with 1.5 hydra

Pavan Balaji balaji at mcs.anl.gov
Sun Dec 23 21:21:57 CST 2012


Rohit,

Can you try the latest mpich version (3.0.1)?  There are some fixes in
there that might be related to this.

 -- Pavan

On 11/06/2012 04:34 PM US Central Time, Jain, Rohit wrote:
> Thanks John. That does give me list of available binding option.
> 
>  
> 
> It seems all binding options don’t work fine.
> 
>  
> 
> All l1/l2/l3cache errors out:
> 
> mpiexec -bind-to l3cache -n 1 mytest 200000 -1 : -n 4 mytest 200000 -1
> 
> [proxy:0:0 at gretel] cache_to_cpu_type
> (/u/prod/mpich/src/mpich2-1.5/src/pm/hydra/tools/topo/hwloc/topo_hwloc.c:258):
> cache level 3 not found
> 
> [proxy:0:0 at gretel] obj_type_to_map_str
> (/u/prod/mpich/src/mpich2-1.5/src/pm/hydra/tools/topo/hwloc/topo_hwloc.c:313):
> error while mapping cache to cpu object
> 
> [mpiexec at gretel] control_cb
> (/u/prod/mpich/src/mpich2-1.5/src/pm/hydra/pm/pmiserv/pmiserv_cb.c:201):
> assert (!closed) failed
> 
> [mpiexec at gretel] HYDT_dmxu_poll_wait_for_event
> (/u/prod/mpich/src/mpich2-1.5/src/pm/hydra/tools/demux/demux_poll.c:77):
> callback returned error status
> 
> [mpiexec at gretel] HYD_pmci_wait_for_completion
> (/u/prod/mpich/src/mpich2-1.5/src/pm/hydra/pm/pmiserv/pmiserv_pmci.c:196):
> error waiting for event
> 
> [mpiexec at gretel] main
> (/u/prod/mpich/src/mpich2-1.5/src/pm/hydra/ui/mpich/mpiexec.c:325):
> process manager error waiting for completion
> 
>  
> 
>  
> 
> This hangs and doesn’t even exit after ‘ctrl-c’:
> 
> mpiexec -bind-to numa -n 1 mytest 200000 -1 : -n 4 mytest 200000 -1
> 
> [mpiexec at gretel] Sending Ctrl-C to processes as requested
> 
> [mpiexec at gretel] Press Ctrl-C again to force abort
> 
> Ctrl-C caught... cleaning up processes
> 
> Ctrl-C caught... cleaning up processes
> 
> Ctrl-C caught... cleaning up processes
> 
> Ctrl-C caught... cleaning up processes
> 
>  
> 
> Here is machine detail I am running it on:
> 
> Processor name                          : Intel(R) Xeon(R) CPU X5550 @
> 2.67GHz
> 
> Speed (MHz)                             : 2660.125
> 
> Cache size per processor                : 8192 KB
> 
> Number of processors                    : 2
> 
> Number of cores per processor           : 4
> 
>  
> 
> Regards,
> 
> Rohit
> 
>  
> 
> *From:*mpich-discuss-bounces at mcs.anl.gov
> [mailto:mpich-discuss-bounces at mcs.anl.gov] *On Behalf Of *John Fettig
> *Sent:* Monday, November 05, 2012 6:05 PM
> *To:* mpich-discuss
> *Subject:* Re: [mpich-discuss] cpu binding with 1.5 hydra
> 
>  
> 
> Rohit,
> 
> Did you try
> 
> $ mpirun -bind-to -help
> 
> This dumps all of the options for -bind-to.
> 
> John
> 
>  
> 
> On Mon, Nov 5, 2012 at 2:13 PM, Jain, Rohit <Rohit_Jain at mentor.com
> <mailto:Rohit_Jain at mentor.com>> wrote:
> 
> Hi Pavan,
> 
> Could you please respond to it?
> 
> Regards,
> Rohit
> 
> -----Original Message-----
> From: mpich-discuss-bounces at mcs.anl.gov
> <mailto:mpich-discuss-bounces at mcs.anl.gov>
> [mailto:mpich-discuss-bounces at mcs.anl.gov
> <mailto:mpich-discuss-bounces at mcs.anl.gov>] On Behalf Of Jain, Rohit
> Sent: Tuesday, October 30, 2012 4:02 PM
> To: Pavan Balaji; mpich-discuss at mcs.anl.gov
> <mailto:mpich-discuss at mcs.anl.gov>
> Subject: Re: [mpich-discuss] cpu binding with 1.5 hydra
> 
> Hi Pavan,
> 
> Where do I find exact names of new binding options arguments?
> 
> Like 'sockets' doesn't work, but 'socket' do. This page is not updated
> with correct options:
> http://wiki.mcs.anl.gov/mpich2/index.php/Using_the_Hydra_Process_Manager#Process-core_Binding
> 
> Regards,
> Rohit
> 
> 
> -----Original Message-----
> From: Pavan Balaji [mailto:balaji at mcs.anl.gov <mailto:balaji at mcs.anl.gov>]
> Sent: Tuesday, October 30, 2012 8:36 AM
> To: mpich-discuss at mcs.anl.gov <mailto:mpich-discuss at mcs.anl.gov>
> Cc: Jain, Rohit
> Subject: Re: [mpich-discuss] cpu binding with 1.5 hydra
> 
> 
> On 10/29/2012 07:13 PM, Jain, Rohit wrote:
>> Do you mean '-binding' option is also being deprecated? Is it replaced
> with '-bind-to'?
> 
> It's being replaced with the -bind-to option.  We might support -binding
> as an alias to -bind-to for quite some time, but we'd suggest users move
> to the new richer -bind-to interface.
> 
>   -- Pvaan
> 
> --
> Pavan Balaji
> http://www.mcs.anl.gov/~balaji
> _______________________________________________
> mpich-discuss mailing list     mpich-discuss at mcs.anl.gov
> <mailto:mpich-discuss at mcs.anl.gov>
> To manage subscription options or unsubscribe:
> https://lists.mcs.anl.gov/mailman/listinfo/mpich-discuss
> _______________________________________________
> mpich-discuss mailing list     mpich-discuss at mcs.anl.gov
> <mailto:mpich-discuss at mcs.anl.gov>
> To manage subscription options or unsubscribe:
> https://lists.mcs.anl.gov/mailman/listinfo/mpich-discuss
> 
>  
> 
> 
> 
> _______________________________________________
> mpich-discuss mailing list     mpich-discuss at mcs.anl.gov
> To manage subscription options or unsubscribe:
> https://lists.mcs.anl.gov/mailman/listinfo/mpich-discuss
> 

-- 
Pavan Balaji
http://www.mcs.anl.gov/~balaji



More information about the discuss mailing list