It is working well to meet my need:

  I set maxujobs to 8 in 'qconf -msconf'
  and maximum slots 8 per any given user
  in RQS. Then I started a bunch of single
  and multiple processor test jobs from
  couple different test user accounts.

  The sum total of slots used per user 
  was always less than or equal to 8 from 
  8 or less jobs while rest of the jobs 
  waited in queue.

Thanks again for your quick help.

Best,
g

--
Gowtham
Information Technology Services
Michigan Technological University

(906) 487/3593
http://www.it.mtu.edu/


On Tue, 5 Jun 2012, Gowtham wrote:

| 
| Thank you, kind Sir. I'll set maxujobs, play around
| with slots some more in RQS and see how that will 
| work.
| 
| Best,
| g
| 
| --
| Gowtham
| Information Technology Services
| Michigan Technological University
| 
| (906) 487/3593
| http://www.it.mtu.edu/
| 
| 
| On Tue, 5 Jun 2012, Reuti wrote:
| 
| | Hi,
| | 
| | Am 05.06.2012 um 16:51 schrieb Gowtham:
| | 
| | > I have read through, to the best of my ability in
| | > last couple days, the Grid Engine archives and
| | > BioTeam's slides on various aspects of SGE, but
| | > I can't seem to get RQS set up to do what I need:
| | > 
| | > 
| | >  Accept as many jobs (serial & parallel) as the
| | >  user submits but run a maximum of 10 jobs per
| | >  user at any given time OR let that user use a
| | >  maximum of 64 slots (used by 10 or less active
| | >  jobs) at any given time
| | > 
| | > 
| | > Here's what I have done so far:
| | > 
| | >  1. qconf -mconf
| | > 
| | >     max_u_jobs 10
| | > 
| | >  2. Added a rule in RQS
| | > 
| | >     {
| | >       name         max_slots_per_user
| | >       description  Limit no. of active slots per user
| | >       enabled      TRUE
| | >       limit        users {*} to slots=64
| | >     }
| | > 
| | > 
| | > Problems I have run into:
| | > 
| | >  1. With #1 (max_u_jobs 10), SGE won't even accept
| | >     the 11th job user submits to the queue
| | 
| | Yes, this limits the number of jobs in the cluster per user, whether they 
are running or waiting. But there is maxujobs in `qconf -msconf` to control the 
running ones.
| | 
| | 
| | >  2. I'm confused a bit with the terms 'slots' and
| | >     'jobs' (BioTeam slides treat slots as jobs).
| | >     To get some clarification, I set max_u_jobs
| | >     to 0 (unlimited) and modified the RQS rule
| | >     to limit the maximum slots to 5.
| | > 
| | >     I then submitted 10 instances of 'Hello, World!'
| | >     (MPI) using 2 processors each and 10 instances
| | >     of 'sleeper.sh' using 1 processor each.
| | > 
| | >     What I noticed is that the number of slots from
| | >     active jobs summed up to 5, irrespective of the
| | >     number of active jobs.
| | 
| | I don't know which particular slide you refer to. By default only "slots", 
i.e. requested "cores" one could say, can be controlled by an RQS.
| | 
| | If you want to control "jobs" by an RQS, you will need to define a complex 
with the value of consumable set to "JOB" (`man complex`) and attach an 
arbitrary high value to `qconf -me global` to have something to consume them 
from.
| | 
| | -- Reuti
| | 
| | 
| | >     So, I do understand now that the term 'jobs' in
| | >     BioTeam slides actually mean 'slots'.
| | > 
| | > 
| | > Help please? Any guidance towards further material
| | > to read up on (man pages, websites, etc.) will be
| | > greatly appreciated.
| | > 
| | > Best,
| | > g
| | > 
| | > --
| | > Gowtham
| | > Information Technology Services
| | > Michigan Technological University
| | > 
| | > (906) 487/3593
| | > http://www.it.mtu.edu/
| | > 
| | > _______________________________________________
| | > users mailing list
| | > [email protected]
| | > https://gridengine.org/mailman/listinfo/users
| | 
| | 
| 
_______________________________________________
users mailing list
[email protected]
https://gridengine.org/mailman/listinfo/users

Reply via email to