[GE users] SGE/6.2 and IBM x3950m2

jbhoren jbhoren at alaska.edu
Thu Feb 4 18:55:51 GMT 2010


    [ The following text is in the "iso-8859-1" character set. ]
    [ Your display is set for the "ISO-8859-10" character set.  ]
    [ Some special characters may be displayed incorrectly. ]

On Thu, Dec 24, 2009 at 2:12 AM, adary <adary at marvell.com<mailto:adary at marvell.com>> wrote:
We use a two node x3950, each one with 128gb ram and 4 quad core CPU's, with RHEL4.7 installed.

Just one single host with 32 CPU's and 256gb ram as far as linux and SGE are concerned

I'm curious about how "smart" the Linux kernel's scheduler is. Specifically, given a job which requires multiple CPUs (really, multiple cores), will it be assigned to CPUs/cores on the same card, or randomly, to available CPUs/cores on separate cards?

Also, a question in resource management: Our users (~250) access an existing cluster (16 nodes, running SGE/6.0u10) via a web-based portal, configured such that all jobs are submitted with the same user name -- "apache".  When we bring our three-node x3950 on-line (and re-purpose the current cluster), access will continue to be via that web-based portal.

Has anyone developed a methodology for allocating resources with a single user?

Given that our users (graduate students, faculty, and researchers) generally work on one project at a time, I thought about using "project" in place of "user". Has anyone done this, or something similar? If so, care to share? Thoughts? Suggestions?


--
JONATHAN B. HOREN
Systems Administrator
UAF Life Science Informatics
Center for Research Services
jbhoren at alaska.edu<mailto:jbhoren at alaska.edu>
http://biotech.inbre.alaska.edu




More information about the gridengine-users mailing list