[GE users] Short medium and long queue setup

Reuti reuti at staff.uni-marburg.de
Wed Oct 18 17:47:10 BST 2006


    [ The following text is in the "ISO-8859-1" character set. ]
    [ Your display is set for the "ISO-8859-10" character set.  ]
    [ Some special characters may be displayed incorrectly. ]

Am 18.10.2006 um 17:45 schrieb Jerome:

> Reuti wrote:
>>>
>>> Hi Jeff,
>>> thank's for your response.
>>> I used the comments you send me to try on our own cluster.
>>> But i notice that the nice value with the mpi jobs just affect  
>>> the  master node of the job. All the other mpich program on the  
>>> "slaves"  node run with a ice value of 0. Do you have the same  
>>> behavior?
>>> Best regards.
>> did you set up the PE with Tight Integration for the parallel  
>> jobs?  All processes on the slaves a children of the sge_shepherd?
> Hi Reuti.
> Yes, i compiled (or i think so) the mpi as it is indicated in the  
> Howto on SGE website. I use the third solution, compilation of the  
> sources.
> If all the process in the paster node of the mpi group are with a  
> nice value of 19, as i show you in teh folowing (command: "ps f -eo  
> nice,user,pgrp,command --cols=70")
>
>   0 sge       3387 /opt/gridengine/bin/lx26-x86/sge_execd
>   0 sge      16942  \_ sge_shepherd-35 -bg
> -19 lorenzo  17017      \_ /bin/tcsh /opt/gridengine/default/spool/c-1
> -19 lorenzo  17017          \_ /bin/sh /opt/mpich/gnu/bin/mpirun -np 8
> -19 lorenzo  17017              \_ /home/lorenzo/mrbayes-3.1.2/mb -p4p
> -19 lorenzo  17017                  \_ /home/lorenzo/mrbayes-3.1.2/mb
> -19 lorenzo  17017                  \_ /opt/gridengine/bin/lx26-x86/qr
> -19 lorenzo  17017                  |   \_ /usr/bin/ssh -n -p 33984 c-
> -19 lorenzo  17017                  \_ /opt/gridengine/bin/lx26-x86/qr
> -19 lorenzo  17017                  |   \_ /usr/bin/ssh -n -p 33985 c-
> -19 lorenzo  17017                  \_ /opt/gridengine/bin/lx26-x86/qr
> -19 lorenzo  17017                  |   \_ /usr/bin/ssh -n -p 34035 c-
> -19 lorenzo  17017                  \_ /opt/gridengine/bin/lx26-x86/qr
> -19 lorenzo  17017                  |   \_ /usr/bin/ssh -n -p 33959 c-
> -19 lorenzo  17017                  \_ /opt/gridengine/bin/lx26-x86/qr
> -19 lorenzo  17017                  |   \_ /usr/bin/ssh -n -p 33947 c-
> -19 lorenzo  17017                  \_ /opt/gridengine/bin/lx26-x86/qr
> -19 lorenzo  17017                  |   \_ /usr/bin/ssh -n -p 33980 c-
> -19 lorenzo  17017                  \_ /opt/gridengine/bin/lx26-x86/qr
> -19 lorenzo  17017                      \_ /usr/bin/ssh -n -p 33891 c-
>
> In one of the slave node, i notice that uniquely the sshd to  
> connect this node from the master one have a nice value of 19:
>
>   0 sge       3364 /opt/gridengine/bin/lx26-x86/sge_execd
>   0 sge      16975  \_ sge_shepherd-35 -bg
> -19 root     16976      \_ sshd: lorenzo [priv]
>   0 lorenzo  16976          \_ sshd: lorenzo at notty

Aha, are you using Debian? Otherwise it might be a problem of the  
sshd by default that the nice value is lost. AFAIK this should be the  
same for all child processes of a niced process. - Reuti


>   0 lorenzo  16983              \_ /opt/gridengine/utilbin/lx26-x86/qr
>   0 lorenzo  17037                  \_ tcsh -c /home/lorenzo/mrbayes-3
>   0 lorenzo  17037                      \_ /home/lorenzo/mrbayes-3.1.2
>   0 lorenzo  17037                          \_ /home/lorenzo/mrbayes-3
>
>
> I will checking if i forget something else in the configuration of  
> SGe and mpi.
> Thank's to your answer.
> Best regards.
>
> -- 
> -- Jérôme
> J'ai une excellente mémoire. Je ne retiens presque rien.
> 	(Georges Perros)
>
> ---------------------------------------------------------------------
> 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