[Swift-devel] precision of jobThrottle in sites.xml

[Allan Espinosa]

Aha.

in the documentation it says  nJobs = (jobThrottle*100)+ 2

Thanks for confirming that.

-Allan

2009/12/7 Michael Wilde <wilde at mcs.anl.gov>:
> Allan, your *throttle value* is off by one.
>
> The formula is: nJobs = (jobThrottle*100)+1
>
> So for 256 jobs you want to set it to 2.55
>
> Mihael me be able to explain the rationale. I think it wasnt designed to be
> directly set by users. But since it so frequently is, perhaps the formula
> should be made simpler, or a nJobs parameter added.
>
> - Mike
>
> On 12/7/09 7:27 PM, Allan Espinosa wrote:
>>
>> I got load balancing off by one. this resulted  in straggler jobs in
>> my workload.
>>
>> sites.xml entry for each site:
>>  <pool handle="BGP_$server">
>>      <execution provider="deef"
>>
>> url="http://$ip:50001/wsrf/services/GenericPortal/core/WS/GPFactoryService"/>
>>      <gridftp url="local://localhost"/>
>>      <workdirectory>$workdir</workdirectory>
>>      <profile namespace="karajan" key="jobThrottle">2.56</profile>
>>      <profile namespace="karajan" key="initialScore">10000</profile>
>>    </pool>
>>
>> job distribution to sites from swift-plot-log:
>> site    JOB_START       JOB_END         APPLICATION_EXCEPTION
>> JOB_CANCELED    unknown         total
>> BGP_000         0       257     0       0       0       257
>> BGP_001         0       257     0       0       0       257
>> BGP_002         0       257     0       0       0       257
>> BGP_003         0       257     0       0       0       257
>> BGP_004         0       257     0       0       0       257
>> BGP_005         0       257     0       0       0       257
>> BGP_006         0       257     0       0       0       257
>> BGP_007         0       241     0       0       0       241
>> BGP_008         0       257     0       0       0       257
>> BGP_009         0       257     0       0       0       257
>> BGP_010         0       257     0       0       0       257
>> BGP_011         0       257     0       0       0       257
>> BGP_012         0       257     0       0       0       257
>> BGP_013         0       257     0       0       0       257
>> BGP_014         0       257     0       0       0       257
>> BGP_015         0       257     0       0       0       257
>>
>> Trying out a jobThrottle of 2.54 doesn't give me a good split either:
>> site    JOB_START       JOB_END         APPLICATION_EXCEPTION
>> JOB_CANCELED    unknown         total
>> BGP_000         0       256     0       0       0       256
>> BGP_001         0       255     0       0       0       255
>> BGP_002         0       255     0       0       0       255
>> BGP_003         4       251     0       0       0       255
>> BGP_004         0       259     0       0       0       259
>> BGP_005         4       251     0       0       0       255
>> BGP_006         0       256     0       0       0       256
>> BGP_007         0       255     0       0       0       255
>> BGP_008         0       255     0       0       0       255
>> BGP_009         0       255     0       0       0       255
>> BGP_010         0       258     0       0       0       258
>> BGP_011         0       255     0       0       0       255
>> BGP_012         0       258     0       0       0       258
>> BGP_013         0       256     0       0       0       256
>> BGP_014         0       255     0       0       0       255
>> BGP_015         0       258     0       0       0       258
>>
>> What do you guys suggest for more precise load distribution? a higher
>> score? try it with jobThrottle=2.55?
>>
>> Thanks,
>> -Allan
>>
>
>



-- 
Allan M. Espinosa <http://allan.88-mph.net/blog>
PhD student, Computer Science
University of Chicago <http://people.cs.uchicago.edu/~aespinosa>