[GE users] How is it possible for a consumable resource to be over-consumed.

Magnus Söderberg magnus.soderberg at switchcore.com
Tue Feb 1 11:29:31 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. ]

Hej.
Gary Thomas wrote:
....
> 
> We have a total of 30 nodes, and when we launch one of our regression runs, we
> sometimes end up in a situation where we have 14 jobs that request mti_lic running at
> the same time.  If I do qstat -f <jobnum> on one of the jobs still waiting in the
> queue, it reports that gc:mti_lice=-6.  How is it possible for a job to get schedule
> when the resource has already reach its limit?
> 
I can't offer an explanation for the behaviour you are seeing since I use a loadsensor 
probing the FlexLM licenseserver for the actual number of licenses available (which works 
great by the way).
What I can offer though is a way to handle the oversubscription. I set up Modelsim to die 
if it doesn't get a license right away (License = noqueue in modelsim.ini) and I have my 
script check why Modelsim quit. If it happens to be a license problem or some other freak 
accident whose symptom will eventually fix itself I do exit 99, telling SGE to resubmit 
the job again. Works like a charm.

My script looks something like this:
-------------------------------------------
vsim <lots of arguments here of course>
$mtiexit = $status
echo "Modelsim done with status $mtiexit"

# check for crasch due to license oversubscription

# exit 4 is licensing problems, 42-45 is also license problems but yet not encounted
if ((${mtiexit} == 4) || ((-f ${mtilogfile}) && (`grep -c "Failure to obtain a Verilog 
simulation license" ${mtilogfile}`)) || \
     (($?cover) && (-f ${mtilogfile}) && (`grep -c "Failure to license for coverage" 
${mtilogfile}`))) then
   echo "license allocation failure"
   exit 99
endif

-- 
Magnus Söderberg, M.Sc.E.E.
Staff Engineer
SwitchCore AB        Phone +46 46 2702560
Emdalavägen 18       Fax   +46 46 2702581
SE-223 69 Lund, Sweden

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