[GE users] queue config question
jwoods at sfwmd.gov
Wed Jan 25 20:52:20 GMT 2006
Is it possible to manage to create a queue with 'n' number of slots
which could be scheduled on any host?
As an example, I have a cluster with 20 nodes. There are a total of 60
CPUs installed, and I have the number of slots per host restricted to be
the same as the number of CPUs. I would like to create three queues,
each with a different number of slots. When a job is submitted, I don't
care which host it is run on, but no more than the max number of slots
for the queue should be scheduled at one time.
I have created a host list which contains all of the execution hosts in
my cluster, and have created the three queues using the host list. The
first queue was created with 50 slots, the second with 15 slots, and the
third with 25 slots. Yes, this exceeds the number of slots that the
hosts will allow -- we don't expect all three queuest to be busy at the
same time, and it would be acceptable for a job to be delayed if there
aren't enough slots available on the hosts.
The problem with my approach is that it creates 50 slots on each queue
instance for the first queue, 15 on each queue instance for the second
queue, and 25 on each queue instance for the third. If I submit 50 jobs
to the second queue (which should only have 15 slots), all of the jobs
get scheduled -- not just the first 15. I would like to see 15 jobs get
scheduled, and the remaining 35 wait for free slots in the queue.
I am either doing something wrong, or am misunderstanding the
documentation. If what I'm doing wrong is not obvious, I can provide
the current configurations for my queues and/or hosts.
More information about the gridengine-users