[GE users] mpiexec/TM (was: BLCR with SGE)
ron_chen_123 at yahoo.com
Wed Feb 23 15:01:58 GMT 2005
And there are more discussions on the LAM devel list,
see this thread:
I agree that SGE is very popular in compute farms, but
getting a cleaner solution for the supercomputer
cluster users running parallel jobs is needed.
I do see a *real* need to get a better integration
with LAM, using scripts seems to be a hack to many
people. Even with MPICH, a common question on this
list is, "is my PE really tight?"
With our current tight PE solution, users need to do
some work to setup a few scripts. This is not totally
out of the box!
Reuti, can you sumarize the current problems, and the
possible solutions you discussed on the LAM devel
Rayson, can you send a list of things that we need to
change/fix in order to get a TM library for SGE,
Please send you guys' findings to the dev list.
(sorry, I don't have time to read all the LAM list and
the mpiexec/LAM source. If you guys can tell me where
we currently are, then may be we can work out
something nicely within a couple of months)
--- Rayson Ho <raysonho at eseenet.com> wrote:
> >I read your mail of course, but it was this one I
> >I looked only on the dev list before.
> Thanks for the pointer :)
> Didn't know that the LAM guys are also waiting for
> this... And I've looked
> at the lam source, they are not making calls to PBS
> directly; but mpiexec
> makes 1 or 2 calls to PBS, so we need a bit more
> Also, I don't think the TM API is going to change
> anytime soon - at least I
> didn't hear anything on the Torque or the PBSPro
Do you Yahoo!?
Take Yahoo! Mail with you! Get it on your mobile phone.
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