[torqueusers] Re: [Mauiusers] Need to partition or allocate nodes to queues?

Garrick Staples garrick at usc.edu
Mon Dec 22 22:13:01 MST 2008

On Mon, Dec 22, 2008 at 03:58:08AM -0500, Jen alleged:
> Hi All,
> I am not sure if this is a torque or Maui question, so please forgive the
> crosspost. :}
> I used to do this blissfully boringly using acl_hosts.  Something has
> changed and this is not working anymore between torque 2.1.x and now.  I
> have the following:
> torque 2.3.5
> maui 3.2.6p21

Why upgrade from 2.1.x?  2.1.x is the most stable branch.  2.3 certainly has some new features, but if you don't need them, I recomment staying with 2.1.x.

> I have a large cluster with several departments and these departments want
> to use the nodes they purchased.  In the past, I set up a queue for each of
> these departments and I assigned nodes to the queues in torque.  Has the way
> to do this changed?  I have seen several messages lately from folks no
> longer able to get it to work...
> I am not married to this method and will happily implement any method that
> will keep my (hundreds) of users separated and on their own nodes.  I do not
> want them to have to change their scripts.  These are heterogeneous, too, so
> it is not just a policy thing - the infiniband jobs won't run on the non IB
> nodes, and the some are 8proc, some are 2, etc.
> I have tried the "neednodes" method.  It did nothing.

I've been using the neednodes method for many years.  I'm currently using
torque 2.1..10 (with some patches up to 2.1.12) and

> I've been trying to do partitions in Maui and that seems to be ignored as
> well.  This is a 64 bit system, by the way.

Can you try a 32bit build of maui?  I've recently noticed some problems with 64bit maui.

> set queue esmg08q resources_max.cput = 10000:00:00
> set queue esmg08q resources_max.ncpus = 179
> set queue esmg08q resources_max.nodect = 22
> set queue esmg08q resources_max.walltime = 10000:00:00
> set queue esmg08q resources_min.cput = 00:00:01
> set queue esmg08q resources_min.ncpus = 1
> set queue esmg08q resources_min.nodect = 1
> set queue esmg08q resources_default.neednodes=altix
> set queue esmg08q resources_min.walltime = 00:00:01
> set queue esmg08q resources_default.cput = 10000:00:00
> set queue esmg08q resources_default.ncpus = 8
> set queue esmg08q resources_default.walltime = 100:00:00

I'd simply this.  ncpus and nodect often don't play well together.

Garrick Staples, GNU/Linux HPCC SysAdmin
University of Southern California

See the Dishonor Roll at http://www.californiansagainsthate.com/

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
Url : http://www.supercluster.org/pipermail/torqueusers/attachments/20081222/01636d71/attachment.bin

More information about the torqueusers mailing list