[GE users] confused about queues
reuti at staff.uni-marburg.de
Mon Feb 7 17:30:36 GMT 2005
Jean-Marc Lachaine wrote:
> I am running a small (8 dual processor nodes) cluster using Rocks and SGE
> 5.3.x I have read the SGE manual, the man pages and have searched the
> archives but I still do not understand things well enough to be able to
> figure out how I would implement 2 queues that would encompass all of the
> available predefined queues that are already set up on each compute node.
> What I would like is to define a long and a short queue. Each queue would
> include, say 4, of the available nodes and would be mutually exclusive. For
> long.q would have access to compute-0-0 to compute-0-3
> short.q would have access to compute-0-4 to compute-0-7
> Each node currently has a compute-0-n.q queue defined each with 2 slots.
> My best guess as to how to accomplish this, based on my understanding of
> what I read is to select one of my execution hosts, say, compute-0-0 and
> define a second queue on it which, based on my example, I would call long.q
> then I subordinate the compute-0-n.q n=(0,...,3) to long.q.
> Does this make sense? Is there something I missed in the manual that
> explains clearly how to crate two queues that would span a total of 8
> execution hosts? Any help would be greatly appreciated.
in 5.3 you have to define in your setup one queue on each compute-0-0 to
compute-0-3 machine and call it e.g. long01 .. long03. The predefined
compute-0-n.q can be deleted. The same you can do with the short queues
short04 ... 08. To avoid getting a job in the wrong queue, you could
define and attach a complex "long" only to the long queues, and request
this (-l long) when you submit the job (otherwise you would have to
specify already the exact queue (-q long02), which would prohibit any
floating scheduling of your jobs between the 4 nodes for each type of
queue; same for short.
Subordinate queues is a feature to suspend queues, i.e. having some
queues with higher priority which suspend lower priority queues.
In 6.0 it's much easier. You define two hostgroups with 4 machines each,
define two cluster queues for each of these hostgroups. And request the
desired cluster queue when you submit the job. I think, this is what you
want in the end.
CU - Reuti
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