[gridengine users] Accounting: wrong CPU time
mazouzi at gmail.com
Tue Feb 4 22:05:37 UTC 2014
Thanks for your replay
It looks like this bug: http://bugs.centos.org/view.php?id=6553
On Tue, Feb 4, 2014 at 9:11 PM, Mazouzi <mazouzi at gmail.com> wrote:
> Hi Reuti,
> I observe this behavior only for certain mutlithreaded applications and
> where the wallclock is few days.
> For the example given above, I've found in the log (the user did time g09
> ...) :
> Starting GAUSSIAN run at Mon Jan 13 21:47:09 CET 2014
> real 17266m2.723s
> user 213746095m54.183s
> sys 93834383m50.758s
> Finished GAUSSIAN run at Sat Jan 25 21:33:12 CET 2
> We can see that values givens by time are also wrong !!
> On Tue, Feb 4, 2014 at 7:06 PM, Reuti <reuti at staff.uni-marburg.de> wrote:
>> Am 04.02.2014 um 18:07 schrieb Mazouzi:
>> > On Tue, Feb 4, 2014 at 5:32 PM, Reuti <reuti at staff.uni-marburg.de>
>> > What was specified in the Gaussian input file for %nprocs= - though it
>> can't be that high?
>> > Hi Reuti,
>> > the value of %nprocs is equal to the requested slots. This problem is
>> noticed only in new installed nodes based on centos 6.
>> I have only access to a cluster with 6.2, there it works like expected
>> AFAICS. Do you also observe this on the command line running just an
>> endless loop while setting a time limit with `-limit -t 60` or so too - it
>> should stop really soon.
>> And does a `time ./loop-for-some-time-app` gives the same (wrong) output
>> about the used up resources? Is there a difference in the bash builtin
>> `time` command and `/usr/bin/time`?
>> -- Reuti
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the users