[gridengine users] Simulation mode ?
michel.hummel at thalesgroup.com
Fri Mar 29 10:06:23 UTC 2013
Thanks for your answers,
I think that, using queues to simulate the exechosts, and jobs doing sleep, I can simulate the design.
[@@ THALES GROUP INTERNAL @@]
De : Reuti [mailto:reuti at staff.uni-marburg.de]
Envoyé : jeudi 28 mars 2013 19:27
À : Jesse Becker
Cc : HUMMEL Michel; users at gridengine.org
Objet : Re: [gridengine users] Simulation mode ?
Am 28.03.2013 um 18:55 schrieb Jesse Becker:
> On Thu, Mar 28, 2013 at 06:34:37PM +0100, Reuti wrote:
>> Am 28.03.2013 um 17:03 schrieb HUMMEL Michel:
>>> I want to integrate OSG in a future system which will manage lot of batch jobs.
>>> I am trying the evaluate the material design, and I want to know of there is a way to simulate the future system with OSG playing a scenario with virtual resources and virtual jobs ?
>> Do you mean, you want to simulate a complete cluster just on one machine where all the jobs request some resources but are doing only a `sleep $RANDOM` in the end?
>> To simulate it with just one running sge_execd, you could setup one queue per virtual host and SGE has to distribute the jobs along these queues. Resources per exechost have then to be added to each queue instead (like "exclusive", "virtual_memory" of a virtual exechost) and not to the exechost definition.
>> If you want to take used CPU time into account for a functional or share tree policy, this would be much harder to simulate of course.
> That's not too hard to do. Just run some simple program that spins
> the CPU and/or eats RAM for a while. Ideally one that can run for
> user-determined period of time (or spin loops, etc).
But the simulation is running only on one machine with a limited number of cores/RAM and not reflecting the simulated cluster in total. Hence the consumed CPU time will be shared to some fraction to each simulated job.
> Something like the 'stress' program:
> Jesse Becker
> NHGRI Linux support (Digicon Contractor)
More information about the users