[GE users] LSF -> SGE .. But there's a wrinkle
andrew at anvil.org
Wed Jun 6 13:55:19 BST 2007
[ 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. ]
I am new to SGE and new to this list, so I hope you will forgive my
piling straight in with a question.
A customer of mine is moving from LSF to SGE and I am to configure SGE
to do much the same as the existing LSF setup which is pretty simple.
There are a dozen or so compute servers, but they are of a variety of
different powers and so cannot be treated as clones of each other. The
jobs to be run are huge. So big that even beefy boxen like these can
only take between 1 and 4 jobs each, depending on the machine.
On LSF we had two parameters called with the job; one for the number of
Gigs of RAM required for the job and the amount of a theoretical CPU
core that would be consumed. Each machine under LSF was then assigned a
number of these theoretical cores .. 3 for a twin dual core, 1 for a
single single core for example. LSF has a relatively sophisticated way
of working out how we're doing wrt memory. If I call a job that need 4
Gigs of RAM and the machine has 6 left, 4 of them get allocated. I am
lead to understand (but have not seen proof) that of the LSF system then
considers this same machine for a 3 Gig job, even if the previous job
has only used part of it's allocated 4G leaving 3G free; the system will
ignore this machine.
I wonder if anyone has any ideas about how to implement this setup in
SGE. I have read through the docs and thought I had something like a
solution, but the jobs just sit there in the queue twiddling their
electrons at me.
Hope someone can help
Andrew Meredith BEng CEng CITP MBCS MIET
The Anvil Organisation Ltd
To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
For additional commands, e-mail: users-help at gridengine.sunsource.net
More information about the gridengine-users