[GE users] execution nodes scalability

Fritz Ferstl ferstl at sun.com
Thu May 18 19:10:44 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. ]

And the Grid Engine master doesn't poll for information. The execds push 
their load data (or job data if they run any) instead. If you've got 
lot's of hosts sending load reports then you might want to set the load 
report interval reasonably high (interval in sec / # of exec hosts = # 
of load reports hitting qmaster on avg per sec ...).

Furthermore, there's no issue with the scheduler, I think. Hosts without 
a queue instance aren't of much interest.

So it should be only the load reporting that could hurt you here.

Cheers,

Fritz

Brooks Davis wrote:
> On Thu, May 18, 2006 at 12:27:43PM -0400, Rayson Ho wrote:
> 
>>I read the paper a long time ago... but if I recall correctly, they
>>only used the command line interface to add/remove nodes, so it should
>>be quite GE version independent. The core part of their stuff is to do
>>load balancing across different GE clusters.
>>
>>However, in your case, you will only need to add/remove nodes when the
>>node comes up, so some simple scripts to call qconf may work nicely.
> 
> 
> FWIW, adding exec hosts is trivial.  I use the following in my script
> to bulk add nodes:
> 
>         qconf -aattr hostgroup hostlist $FQDN @allhosts
>         qconf -as ${FQDN}
>         qconf -ah ${FQDN}
> 
>         mkdir -p ${SGE_ROOT}/${SGE_CELL}/spool/${HOST}
>         mkdir -p ${SGE_ROOT}/${SGE_CELL}/spool/${HOST}/active_jobs
>         mkdir -p ${SGE_ROOT}/${SGE_CELL}/spool/${HOST}/jobs
>         mkdir -p ${SGE_ROOT}/${SGE_CELL}/spool/${HOST}/job_scripts
>         chown -R sgeadmin ${SGE_ROOT}/${SGE_CELL}/spool/${HOST}
> 
> Other than installing a startup script (that I don't use becuase the
> FreeBSD port installed a better one), that's all inst_execd appears
> to do.  I'd probably just keep the directories around (2^16 * 4 extra
> directories isn't that big a deal, just avoid typing ls without -f
> (don't sort) in ${SGE_ROOT}/${SGE_CELL}/spool.  Alternatly you could use
> local spool dirs and avoid the issue entierly.
> 
> -- Brooks
> 
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
> For additional commands, e-mail: users-help at gridengine.sunsource.net
> 



    [ Part 2: "Attached Text" ]

---------------------------------------------------------------------
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