On 29/05/12 16:11, Rayson Ho wrote: > BTW, since you mentioned that you are using our hwloc drop in > replacement for SGE 6.2u5, I just want to make sure that you are using > the execd & shepherd pair.
I downloaded http://gridscheduler.sourceforge.net/projects/hwloc/sge-hwloc-ssl.tar.gz and replaced: /usr/lib/gridengine/sge_execd /usr/lib/gridengine/sge_shepherd /usr/lib/gridengine/loadcheck with the files from the archive. Those are the debianised locations for those binaries. We have dual Magny-Cours based Opteron 6174s on each node so with the upgraded binaries loadcheck output seems to be aware of the NUMA layout of those cores, where it had no idea before: $ /usr/lib/gridengine/loadcheck arch lx26-amd64 num_proc 24 m_socket 2 m_core 24 m_topology SCCCCCCCCCCCCSCCCCCCCCCCCC load_short 1.55 load_medium 2.43 load_long 4.00 mem_free 38429.750000M swap_free 49088.675781M virtual_free 87518.425781M mem_total 48270.757812M swap_total 49143.996094M virtual_total 97414.753906M mem_used 9841.007812M swap_used 55.320312M virtual_used 9896.328125M cpu 3.7% I'm not sure whether that actually translates into better performance or not. Most of our workloads are genomic analyses which means slices of large chunks of data, each being analysed by separate processes. Cheers, Rob -- Robert Hutton Senior Systems and Database Administrator Centre for Genomics and Global Health <http://cggh.org> The Wellcome Trust Centre for Human Genetics Roosevelt Drive Oxford OX3 7BN United Kingdom Tel: +44 (0)1865 287721 _______________________________________________ users mailing list [email protected] https://gridengine.org/mailman/listinfo/users
