<meta http-equiv="Content-Type" content="text/html; charset=utf-8"><div dir="ltr"><div><div>Ken,<br><br></div>"-bind-to core" gives me the following topology:<br><br>process 0 binding: 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 <br>process 1 binding: 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 <br>process 2 binding: 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 <br>process 3 binding: 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 <br>process 4 binding: 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 <br>process 5 binding: 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 <br>process 6 binding: 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 <br>process 7 binding: 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 <br>process 8 binding: 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 <br>process 9 binding: 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 <br>process 10 binding: 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 <br>process 11 binding: 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 <br>process 12 binding: 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 <br>process 13 binding: 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 <br>process 14 binding: 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 <br>process 15 binding: 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0<br><br></div>but I want this:<br><br>process 0 binding: 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 <br>process 1 binding: 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 <br>process 2 binding: 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 <br>process 3 binding: 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 <br>process 4 binding: 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 <br>process 5 binding: 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 <br>process 6 binding: 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 <br>process 7 binding: 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 <br>process 8 binding: 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 <br>process 9 binding: 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 <br>process 10 binding: 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 <br>process 11 binding: 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 <br>process 12 binding: 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 <br>process 13 binding: 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 <br>process 14 binding: 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 <br>process 15 binding: 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0<br><div><div><br></div><div>The latter gives me better performance for my application, and I am guessing it's because I have evenly distribute the processes among the two sockets (sequentially). Which is why I resorted to what I had originally with the custom binding.<br><br></div><div>Thanks,<br></div></div></div><div class="gmail_extra"><br><div class="gmail_quote">On Mon, May 11, 2015 at 4:53 PM, Kenneth Raffenetti <span dir="ltr"><<a href="mailto:raffenet@mcs.anl.gov" target="_blank">raffenet@mcs.anl.gov</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">Justin,<br>
<br>
Try using the "-bind-to core" option instead. It should do exactly what you are wanting. See this page with examples for more details <a href="https://wiki.mpich.org/mpich/index.php/Using_the_Hydra_Process_Manager#Process-core_Binding" target="_blank">https://wiki.mpich.org/mpich/index.php/Using_the_Hydra_Process_Manager#Process-core_Binding</a><br>
<br>
Ken<div><div class="h5"><br>
<br>
On 05/11/2015 04:48 PM, Justin Chang wrote:<br>
</div></div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div><div class="h5">
Hello everyone,<br>
<br>
I am working with an HPC machine that has this configuring for a single<br>
compute node:<br>
<br>
Machine (64GB total)<br>
   NUMANode L#0 (P#0 32GB)<br>
     Socket L#0 + L3 L#0 (25MB)<br>
       L2 L#0 (256KB) + L1d L#0 (32KB) + L1i L#0 (32KB) + Core L#0 + PU<br>
L#0 (P#0)<br>
       L2 L#1 (256KB) + L1d L#1 (32KB) + L1i L#1 (32KB) + Core L#1 + PU<br>
L#1 (P#1)<br>
       L2 L#2 (256KB) + L1d L#2 (32KB) + L1i L#2 (32KB) + Core L#2 + PU<br>
L#2 (P#2)<br>
       L2 L#3 (256KB) + L1d L#3 (32KB) + L1i L#3 (32KB) + Core L#3 + PU<br>
L#3 (P#3)<br>
       L2 L#4 (256KB) + L1d L#4 (32KB) + L1i L#4 (32KB) + Core L#4 + PU<br>
L#4 (P#4)<br>
       L2 L#5 (256KB) + L1d L#5 (32KB) + L1i L#5 (32KB) + Core L#5 + PU<br>
L#5 (P#5)<br>
       L2 L#6 (256KB) + L1d L#6 (32KB) + L1i L#6 (32KB) + Core L#6 + PU<br>
L#6 (P#6)<br>
       L2 L#7 (256KB) + L1d L#7 (32KB) + L1i L#7 (32KB) + Core L#7 + PU<br>
L#7 (P#7)<br>
       L2 L#8 (256KB) + L1d L#8 (32KB) + L1i L#8 (32KB) + Core L#8 + PU<br>
L#8 (P#8)<br>
       L2 L#9 (256KB) + L1d L#9 (32KB) + L1i L#9 (32KB) + Core L#9 + PU<br>
L#9 (P#9)<br>
     HostBridge L#0<br>
       PCIBridge<br>
         PCI 1000:0087<br>
           Block L#0 "sda"<br>
       PCIBridge<br>
         PCI 15b3:1003<br>
           Net L#1 "eth0"<br>
           Net L#2 "ib0"<br>
           OpenFabrics L#3 "mlx4_0"<br>
       PCIBridge<br>
         PCI 8086:1521<br>
           Net L#4 "eth1"<br>
         PCI 8086:1521<br>
           Net L#5 "eth2"<br>
       PCIBridge<br>
         PCI 102b:0533<br>
       PCI 8086:1d02<br>
   NUMANode L#1 (P#1 32GB) + Socket L#1 + L3 L#1 (25MB)<br>
     L2 L#10 (256KB) + L1d L#10 (32KB) + L1i L#10 (32KB) + Core L#10 +<br>
PU L#10 (P#10)<br>
     L2 L#11 (256KB) + L1d L#11 (32KB) + L1i L#11 (32KB) + Core L#11 +<br>
PU L#11 (P#11)<br>
     L2 L#12 (256KB) + L1d L#12 (32KB) + L1i L#12 (32KB) + Core L#12 +<br>
PU L#12 (P#12)<br>
     L2 L#13 (256KB) + L1d L#13 (32KB) + L1i L#13 (32KB) + Core L#13 +<br>
PU L#13 (P#13)<br>
     L2 L#14 (256KB) + L1d L#14 (32KB) + L1i L#14 (32KB) + Core L#14 +<br>
PU L#14 (P#14)<br>
     L2 L#15 (256KB) + L1d L#15 (32KB) + L1i L#15 (32KB) + Core L#15 +<br>
PU L#15 (P#15)<br>
     L2 L#16 (256KB) + L1d L#16 (32KB) + L1i L#16 (32KB) + Core L#16 +<br>
PU L#16 (P#16)<br>
     L2 L#17 (256KB) + L1d L#17 (32KB) + L1i L#17 (32KB) + Core L#17 +<br>
PU L#17 (P#17)<br>
     L2 L#18 (256KB) + L1d L#18 (32KB) + L1i L#18 (32KB) + Core L#18 +<br>
PU L#18 (P#18)<br>
     L2 L#19 (256KB) + L1d L#19 (32KB) + L1i L#19 (32KB) + Core L#19 +<br>
PU L#19 (P#19)<br>
<br>
If I ran my program with 16 processes, I would have the follow batch script:<br>
<br>
#!/bin/bash<br>
#SBATCH -N 1<br>
#SBATCH -n 20<br>
#SBATCH -t 0-09:00<br>
#SBATCH -o output.txt<br>
<br>
mpiexec.hydra -bind-to user:0,1,2,3,4,5,6,7,10,11,12,13,14,15,16,17 -n<br>
16 ./my_program <args><br>
<br>
This would give me decent speedup. However, what if I want to use 32<br>
processes? Since each node only has 20 cores I would need #SBATCH -N 2<br>
and #SBATCH -n 40. However, I want ranks 0-15 and 16-31 to have the same<br>
mapping as above but on different compute nodes, so how would I do this?<br>
Or would the above line work so long as I have a multiple of 16 processes?<br>
<br>
Thanks,<br>
<br>
--<br>
Justin Chang<br>
PhD Candidate, Civil Engineering - Computational Sciences<br>
University of Houston, Department of Civil and Environmental Engineering<br>
Houston, TX 77004<br>
<a href="tel:%28512%29%20963-3262" value="+15129633262" target="_blank">(512) 963-3262</a><br>
<br>
<br></div></div>
_______________________________________________<br>
discuss mailing list     <a href="mailto:discuss@mpich.org" target="_blank">discuss@mpich.org</a><br>
To manage subscription options or unsubscribe:<br>
<a href="https://lists.mpich.org/mailman/listinfo/discuss" target="_blank">https://lists.mpich.org/mailman/listinfo/discuss</a><br>
<br>
</blockquote>
_______________________________________________<br>
discuss mailing list     <a href="mailto:discuss@mpich.org" target="_blank">discuss@mpich.org</a><br>
To manage subscription options or unsubscribe:<br>
<a href="https://lists.mpich.org/mailman/listinfo/discuss" target="_blank">https://lists.mpich.org/mailman/listinfo/discuss</a><br>
</blockquote></div><br></div><br clear="all"><br>-- <br><div class="gmail_signature"><div dir="ltr"><div><div><div>Justin Chang<br></div>PhD Candidate, Civil Engineering - Computational Sciences<br></div>University of Houston, Department of Civil and Environmental Engineering<br></div>Houston, TX 77004<br>(512) 963-3262<br></div></div>