[GE users] resource allocation and race condition

Magnus Söderberg magnus.soderberg at switchcore.com
Tue Oct 19 10:11:07 BST 2004


    [ The following text is in the "ISO-8859-15" character set. ]
    [ Your display is set for the "ISO-8859-10" character set.  ]
    [ Some special characters may be displayed incorrectly. ]

The only "real" solution to oversubscription would be if FlexLM would allow reservation of 
a license prior to a job starting or implementing such a mechanism through the use of a 
proxy license server, as someone have mentioned before on this list.

With that said, wouldn't it be possible to instead of implementing the perfect license 
tracker simply resubmit jobs that fail due to license errors?
We have a load sensor for FlexLM licenses as explained on Chris Dagdigians web page and 
the interactive users uses the same licenses as is used through SGE (and on the same hosts 
as well).
I run Modelsim through SGE and for me it works great, the few times a Modelsim job is 
started and doesn't get a license due to delays in the load sensor, I simply detect that 
at the end of my job script and exit the script with status 99, telling SGE to rerun the 
job. An explicit call to qresub, possibly with increasing the priority would also work I 
guess.
This is dependent on the possibility to tell Modelsim to die immediately if it doesn't get 
all the license tokens needed. However that isn't a particularly high-tech feature and any 
vendor could easily implement it if not already in the software.

regards

Magnus Söderberg

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