[GE users] question about queues and limitations to specific nodes
reuti at staff.uni-marburg.de
Wed May 12 22:20:48 BST 2010
Am 12.05.2010 um 22:28 schrieb jnorris:
> Here is the situation:
> I have 66 nodes total in my cluster - I need to allow select users to
> access the first 8 nodes and all users access the remaining nodes.
> The first 8 nodes I put into a queue ib.q and the rest in
> ethernet.q I
> set the exclude user list on IB to all of the user lists of those that
> should not be allowed and the user list to the users that are allowed.
is the list of the "extra-users" a subset of the "users" one? When a
user is in both, access will be denied.
> With ethernet.q I have all user list in users and NONE for exclude
> Then I did some testing with a simple hello script that keeps track of
> what node its running on and prints the message:
> Hello from process 0 on c1
> Hello from process 1 on c2
> Hello from process 2 on c3
> Hello from process 3 on c4
> Hello from process 4 on c5
> Hello from process 5 on c6
> Hello from process 6 on c7
> Hello from process 7 on c8
> I set the #$ -q ethernet.q in the script and set the run as:
> mpiexec.hydra -f /opt/mpdd_tmp/hydra.hosts -n 10 ./hello
Is Hydra now final? When I checked it last, it was only a beta and I
didn't look further into it. AFAIK Hydra is not tightly integrated
So we would need to use rsh/ssh to be directed to a `qrsh -inherit`.
For now it looks like it's using any node and not the granted ones.
Which type of PE did you set up?
> The out put above is what I obtain. Ok here are my questions:
> 1) is this the best way to do this? - I am real newbie to sge
In some way this is also a personal taste how you would like to -
let's say - visualize the usage. In principle you can also have:
- only one queue
- one hostgroup @ib_nodes for the first 8 nodes (to avoid specifying
each machine [see two lines below])
- one users list called "others" which contains only the ordinary
users w/o the special ones
- set in this queue: xuser_lists none,[@ib_nodes=others]
One more pitfall: when you attach the same PE to two queues you can
also get a mixture of slots from both queues in your setup. A second
PE hydra.ib, which is attached to
pe_list hydra,[@ib_nodes=hydra hydra.ib]
(or "pe_list hydra,[@ib_nodes=hydra.ib]" if ethernet jobs shouldn't
run on the IB nodes at all)
some nodes, can help. And submit either "-pe hydra 4" or "-pe hydra.ib
4" to choose the machines. Advantage: use "-pe hydra* 4" if you simply
don't care what you get.
> 2) by the scenario above and considering that c1-c8 are the nodes
> should not be in this output - what am I doing wrong?
> Thank beforehand for any and all help.
> Joseph Norris
> Application Developer & Server Administrator
> jnorris at ucmerced.edu
> To unsubscribe from this discussion, e-mail: [users-unsubscribe at gridengine.sunsource.net
To unsubscribe from this discussion, e-mail: [users-unsubscribe at gridengine.sunsource.net].
More information about the gridengine-users