[GE users] qhost show wrong NCPU value

Reuti reuti at staff.uni-marburg.de
Fri May 19 10:25:31 BST 2006


Hi,

Am 18.05.2006 um 22:17 schrieb Thiep Duong:

> I just upgrade my SGE from 5.3 to 6.0u7 recently. Things works fine
> for me so far.  When I do qhost, I got some of the number under NCPU
> reported '0' or negative' number.
>
> I am making a user's defined consumable complex with name num_proc --
> To allow user to fill up the slots on a system (l slots=2 is not  
> allowed).
> Is this num_proc is a reserved word on SGE_6.0?
>
> %qhost
> HOSTNAME                ARCH         NCPU  LOAD  MEMTOT  MEMUSE   
> SWAPTO  SWAPUS
> ---------------------------------------------------------------------- 
> ---------
> global                  -               -     -       -        
> -       -       -
> blinux1                 lx24-x86        2  0.00    3.8G  245.3M    
> 17.7G   44.5M
> blinux2                 lx24-x86        2  0.01    3.8G  251.8M    
> 17.7G   36.5M
> blinux3                 lx24-x86        2  0.00    3.8G  252.6M    
> 18.0G   33.8M
> blinux5                 lx24-x86        2  0.00    3.8G  271.5M    
> 18.0G     0.0
> optqlinux19             lx24-amd64      4  0.00   30.8G  511.0M   
> 136.7G   29.6M
> optlinux10              lx24-amd64      2  0.00   15.4G  730.2M    
> 46.2G   16.1M
> blinux6                 lx24-x86        0  2.00    3.7G    1.3G    
> 17.8G   30.6M
> scblad03                sol-sparc64    -2  0.01   16.0G    2.3G    
> 26.1G   11.0M
>
> Problem only seems to be on older Redhat-7/8 system, or some  
> solaris system.
> I don't see problem on any of my opteron, running redhat-3.0
>
> Any better suggestion for user to reserve the whole system to run job
> other than mimicking the consumable slots?

this was already on the list before - just search for "exclusive" or  
so. In short my way to do it:

- define a PE for each platform with a fixed allocation rule of no.  
of CPUs in this platform
- give them similar names, which can be requested with a wildcard
- submit also serial jobs with "-pe excl* 2-99 - R y" (set  
max_reservation 20 in the scheduler configuration [or to a value more  
appropriate for your job-turnaround])

on whatever platform you end up, you will have all slots inside this  
machine.

(Other way maybe to request a consumable resource like virtual_free  
and simply specify the complete amount in the machine, but this will  
work only in uniform clusters, or if you need to end up on one  
specific platform anyway).

HTH - Reuti


> Thanks.
>
> Thiep.Duong at am.necel.com
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
> For additional commands, e-mail: users-help at gridengine.sunsource.net

---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
For additional commands, e-mail: users-help at gridengine.sunsource.net




More information about the gridengine-users mailing list