[torqueusers] NUMA config question on build from trunk.

Mike Coyne Mike.Coyne at PACCAR.com
Mon Oct 11 10:22:00 MDT 2010


I am working with a ScaleMP linux system  which consists of 10 dual
socket blades which creates 10 nodes 0-19. due to the cpu speed on 4 of
the blades they show up a nodes with memory but no cpus , 

My mom layout 

cpus=0-11  mem=0,1 memsize=441738948

cpus=12-23  mem=2,3 memsize=441738948

cpus=23-35  mem=4,5 memsize=441738948

cpus=36-47  mem=6,7 memsize=441738948

cpus=48-59  mem=8,9 memsize=441738948

cpus=60-71  mem=10,11 memsize=441738948

 

for the first 6 baldes , they show up as <hostname>-0...5 in pbsnodes
and in moab's mdiag -n  I would like to  be able to submit to a job
locked to a "blade" which can use memory from other/(the additional mems
12-19) in the preferred allocation strategy, getting  memory from the
local node first , then from others.  How do I tell moab/torque that the
available memory is the "SUM" of the memory on all the nodes + the extra
non-cpu nodes not the available memory each individual node?  And how do
I configure in nodes 12-19 I tried adding them to the mem=-0.1,12,13...
lines but I don't want them to be preferred per-say .

 

 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.supercluster.org/pipermail/torqueusers/attachments/20101011/1725ae4d/attachment.html 


More information about the torqueusers mailing list