Re: [Condor-users] Questions about ImageSize

[Lukas Slebodnik <slebodnik@xxxxxxxx>]

Thanky you for quick reply. It help me a lot.

On Thu, Feb 03, 2011 at 07:55:45AM -0500, Matthew Farrellee wrote:
> On 02/03/2011 04:36 AM, Lukas Slebodnik wrote:
>> Hi All,
>>
>>     I have two questions:
>> I can see in output od fommand condor_q -l ImageSize and ImageSize_RAW. It seems
>> that ImageSize is rounded. (Why?)
>> #$condor_q -l | grep ^ImageSize
>> #ImageSize_RAW = 1343716
>> #ImageSize = 1500000
>> #ImageSize_RAW = 1311296
>> #ImageSize = 1500000
>>
>> And which one should be used in SYSTEM_PERIODIC_HOLD expression.
>
> Condor contains optimizations that automatically group jobs together.  
> The grouping uses any attributes that appear in START expressions, which  
> sometimes includes ImageSize. The rounding is to create fewer levels for  
> the ImageSize variable and thus fewer groups.
>
> You may want to use ImageSize_RAW.

SYSTEM_PERIODIC_HOLD = \
    (JobStatus == 1 || JobStatus == 2) && \
    (JobRunCount > 10 || ImageSize > (RequestMemory * 1024))

I try both ImageSize and ImageSize_RAW in previous expression and only
ImageSize work properly. I also tried some modification with ifThenElse(), but
this didn't help. Only one solution is to use "ImageSize".

I want to edit job requirements (using condor_qedit) for held job and then
release job(using condor_release). So this was the reason for using
ImageSize_RAW for minimization dynamic slots memory usage.

Appropriate solution for me is to modify ImageSize rounding. Default value is
SCHEDD_ROUND_ATTR_ImageSize = 25%
And I change it to round up to 4 decimal places. (10000 KiB)
SCHEDD_ROUND_ATTR_ImageSize = 4

>> Next question is also about ImageSize :).
>> I set up STARTER_UPDATE_INTERVAL = 120, for faster ClassAd updates, that the
>> condor_starter daemon sends to other daemons. Default is 5 minutes. I submit
>> simple job and I can see in log file, that ImageSize are peridically updated.
>> But in "condor_q" there is SIZE 0.0 . I also try out "condor_q -direct rdbms"
>> and "condor_q -direct schedd", but result is the same.
>>
>> After 16 minutes I can see correct SIZE in condor_q output.
>>
>> I would be very gratefull for any comments.
>
> http://spinningmatt.wordpress.com/2009/04/11/publishing-rates-in-a-condor-pool/
>
> STARTER_UPDATE_INTERVAL controls starter -> shadow update.  
> SHADOW_QUEUE_UPDATE_INTERVAL (15min) controls shadow -> schedd update.
>
> So you'll have to change both.
>
>
> Best,
>
>
> matt
>