Hi,
Am 27.10.2011 um 18:22 schrieb He, Janet (NIH/NIAID) [C]:
> 1. What is the purpose of all.q in SGE? What is the optimize way to
> configure all.q?
it's just a queue which is created for your convenience during installation.
There is nothing special about it, and instead of being named "all.q" it could
also be called "foobar", "demo" or whatever. Even the suffix .q is only a
commentary part. And if you don't need the queue, you can remove it completely.
> Should all.q have all the cores in the cluster or a subset of the cores?
> Since the host list for all.q includes all the nodes in the cluster, when
> jobs submit to this queue, how to manage the oversubscribtion for those nodes
> that also belong to other queues?
You can either define the total number of slots per machine in the exechost
definition (complex_values slots=N), or by an RQS (limit hosts {*} to slots=N).
While the first is a little bit more work to have a loop to set it for all
machines (and the latter one is just one line for all), the latter might
clutter the output of `qquota`. But in the end it's personal taste which way
you prefer.
> 2. Is there a way to assign node to multiple queues but not make the node
> over subscribed? Are there any tips or reference site with the information?
See above.
> 3. How exactly np_load_avg used in the job scheduing? Is this applying to
> all the nodes in the queue? How do I calculate the value for np_load_avg if
> the nodes specification in the queue are not the same?
The np_load_avg will only be used for serial jobs to determine on which node a
job should be scheduled to. This is the setting in "queue_sort_method" in the
scheduler (man sched_conf). For parallel ones the "allocation_rule" setting
will be used in the PE definiton.
> 4. If a node is hyper-threading enabled, should we count the real core
> number in SGE or using the hyper-threading core number in SGE, i.e for 8-core
> node, with hyper-threading, the core becomes 16. Should SGE uses 8 or 16
> in the configuration?
This depends on the application. I found by some test runs that Hyperthreading
will give around 1.5 times the performance for our applications but won't
double it. To be on the safe side also for other applications, I turn
Hyperthreading always off and use the real number of cores.
-- Reuti
_______________________________________________
users mailing list
[email protected]
https://gridengine.org/mailman/listinfo/users
