[torqueusers] random node selection using maui?
wyckoff at yahoo-inc.com
Thu Feb 8 15:29:13 MST 2007
the randomization goal is to have the computation on as many different
physical racks as possible. This is for data locality (i.e., latency)
and IO bandwidth scaling for a distributed file system.
The other way we are looking at is, just like you proposed, having a
node property called rackid-XXXX where XXXX = the subnet and then when
doing qsubs, requesting N/#racks of each rack type as a scheduling
I was thinking that if there were a way to be completely random (pseudo
random :)), we could get the same effect.
Lennart Karlsson wrote:
> wyckoff at yahoo-inc.com said:
>> I have a largish cluster (~3K nodes) and I want the ability to
>> allocate a random N nodes on the cluster. The maui documentation
>> says that FASTEST allocation with no attributes is random, but it
>> isn't. Is there another way to do this?
> As far as I know, you have to do the randomization yourself. The first
> three ways that comes to my mind is:
> 1/ Change the Maui code to add a randomization parameter. The hard way.
> 2/ Stop Maui, stop Torque, randomize the node ordering in the Torque
> configuration file "nodes", start Torque, and start Maui, in that
> sequence. A lot of restarts of the applications so it is not at all
> practical if you need to do it more than a few times. But you have not
> told us why you want the randomization: If you actually just want to
> schedule the nodes in a different order, you can reorder the "nodes"
> file once in whatever way you want and be happy ever after. :-)
> 3/ Choose a number of random node names from your cluster and give
> those node names as qsub parameters, e.g.
> qsub -l nodes=n125+n2836+n1583 scriptfile
> Best wishes,
> -- Lennart Karlsson <Lennart.Karlsson at nsc.liu.se>
> National Supercomputer Centre in Linkoping, Sweden
> torqueusers mailing list
> torqueusers at supercluster.org
More information about the torqueusers