[torqueusers] question about default fifo scheduler

David Singleton David.Singleton at anu.edu.au
Tue Jan 31 17:47:13 MST 2006


Garrick Staples wrote:
> On Tue, Jan 31, 2006 at 10:28:21AM -0800, Alexander Saydakov alleged:
> 
>>Hi!
>>
>>Does anyone know how does the default fifo scheduler determine the
>>availability of the nodes in is_ok_to_run_job()? Particularly I would like
>>to know if a node (cpu) can become available in a middle of a scheduling
>>cycle?
>>
>> 
>>
>>My reasoning is as follows: A scheduling cycle begins (in my configuration)
>>with sorting of queues according to priority and sorting of jobs within
>>queues according to priority. Then the scheduler considers those sorted jobs
>>one by one and calls is_ok_to_run_job(). If a free slot can become available
>>any time during this process, that means any job can happen to run
>>regardless of priority. Is this possible? I think that it is what I see. If
>>so how Maui does that?
> 
> 
> I'm not sure about pbs_sched, but it is not possible in maui/moab.  They
> query pbs_server for all job info, all node info, and then make
> scheduling decisions on that information.  There are no repeated queries
> during a scheduling cycle.
> 
> I'd assume pbs_sched works the same way, but someone else will have to
> check the code to confirm.
> 
> 

Garrick is correct.  All information is gathered from the server at
the start of a scheduling cycle.

-- 
--------------------------------------------------------------------------
    Dr David Singleton               ANU Supercomputer Facility
    HPC Systems Manager              and APAC National Facility
    David.Singleton at anu.edu.au       Leonard Huxley Bldg (No. 56)
    Phone: +61 2 6125 4389           Australian National University
    Fax:   +61 2 6125 8199           Canberra, ACT, 0200, Australia
--------------------------------------------------------------------------


More information about the torqueusers mailing list