[GE users] SGE-6.0u6: job_load_adjustments for licenses?

Chris Dagdigian dag at sonsorol.org
Mon Nov 14 17:54:24 GMT 2005


    [ 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. ]

You need to post more information about "why it doesnt work" so we  
can try to figure out what is going on.

What do your load sensors report, What is the configuration of those  
values in the SGE Complex, what are your global settings and host  
settings regarding these values? etc. etc.

Generally speaking, consumable values in SGE should handle the "many  
jobs submitted at once" condition. The race conditions come from jobs  
that take a while to start working after they have been dispatched --  
the race condition here is that a new loadsensor interval kicks in,  
refreshes the internal SGE counts and that causes more jobs to go out  
which then die due to license errors.

Mark Oleson came up with a great method that removes most if not all  
of the most serious race conditions.

He wrote *about* the methods involved in posts that are available here:

http://gridengine.info/articles/2005/09/30/flexlm-licensing-and-grid- 
engine-a-new-howto-draft.

The method involves directly altering the values rather than waiting  
for the polling mechanisms to kick in.

Mark put his words into practice with a great perl license management  
daemon called "qlicserver" - I've been running it for a week or two  
now and it works great. I can publish his code and support it on this  
list as soon as Mark decides upon an appropriate license.  His code  
does assume FlexLM is the license server so if that is not the case  
for you, some additional coding may be required. The "direct  
manipulation of consumables" method is fantastic though so it is  
worth the effort.

-Chris


On Nov 14, 2005, at 12:45 PM, Erik Soyez wrote:

> Hi,
>
> we use Load-Sensors to report free licenses, but if many jobs are
> submittet at the same time, we get job errors due to "race  
> conditions".
>
> Now I would like to use the "job_load_adjustments" feature for  
> licenses,
> i.e. artificially count down the licenses for each "-l license...."
> statement for e.g. 1 minute.
>
> Possible?  Why doesn?t it work in my test setup?
>
> Many thanks,
>
> Erik.
>
> --
>
>
>
> ---------------------------------------------------------------------
> 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