[GE users] SGE Job files and directories

Andy Schwierskott andy.schwierskott at sun.com
Tue Oct 31 17:07:20 GMT 2006


Hi,

to reduce spooling overhead in classic spooling we introduced the creation
of individual files for the task of an array job - for this it was necessary
to create directories under the jobs/ directory.

In Unix hover a directory cannot hold more than 2^16 subdirectories :-(.
This made it necessary to create a more complex directory structure that a
site can have more than 2^16 running and pending jobs.

What you observe is just an internal implementation detail which is of
course completely solved wiht BDB spooling. It does not reflect parallel
jobs.

Regards,
(SGE-)Andy

> Hi SGE,
>
> I noticed that for jobs submitted to more than one node/slot, there is a
> directory created as opposed to jobs submitted to only one node/slot.
> Under the following directory path on the head node.
> /opt/gridengine/default/spool/qmaster/jobs/00/0000
>
> Curious to know, What is the idea behind this ?
>
> Also the "common" file in the directories created differ from the "common"
> file under the sub-directory
> #diff /opt/gridengine/default/spool/qmaster/jobs/00/0000/0841/common
> /opt/gridengine/default/spool/qmaster/jobs/00/0000/0841/1-4096/1/common
>
> Are these files feeding information to Grid Engine specific to individual
> hosts ?
>
> Thank you for any comments,
> -AK

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