Choose timezone
Your profile timezone:
WLCG Accounting Task Force Meeting
→
Europe/Zurich
513/R-068 (CERN)
Attended by:
Dimitrios, Ivan, Gavin, Costin, Greg, John, Alessandra, Adrian, Julia
Julia mentined that there was no much input from the experiments regarding their requirements for the additional attributes to be included in the APEL dataset. Alessandro sent input for ATLAS, mainly insisting of decoupling raw wallclock time from scaled wallclock time, both in the reports and UI. Would be also nice to include some metrics which contains experiment measurements regarding performance.
Since apparently there is no full understanding which additional metrics we want in the APEL dataset apart of raw wallclock decoupled from scaled wallclock, Julia suggested to concentrate first on raw wallclock metric. John's presentation was describing possible implementation. People agreed that option 2 from John's presentation looks to be the best option, though Alessandra suggested to work however on modification of Apel client so that it would be able to properly handle report with per job duration, since it would allow to save time on the server where otherwise wallclock has to be calculated based on start time and endtime time stamps.
For the smry reports changes, Apel needs to perform modifications (client, schema, etc...). When these changes are done we should ask sites which send smry records to add sum of job duration which corresponds to raw wallclock to update their smry generation procedures to include this metric. Apparently there are not too many sites to be contacted, since US sites are not reporting individually to APEL but via Gratia. Gavin confirmed that for CERN the change is possible and not too difficult to perform.
Julia asked whether at some point APEL client can use information from MJF. Apparently not, since as Alessandra explained, APEL client does not communicate with the WN, it communicates with CE.
Julia asked at what point APEL client communicates with BDII. John explained that it is done when usage record is generated.
------
The second topic was the report about storage space accounting work in APEL.
Julia asked why there is such a huge discrepancy of the numbers shown in the prototype compared to ATLAS space accounting in Dashboard. The difference is ~3 orders of magnitude. John told that this has to be investigated.
Julia asked whether APEL client for storage space accounting has any dependencies on BDII, John said that it does not.
Julia asked whether storage space accounting can handle finer granularity than VO storage share at the site. John answered that STAR record allows this, but this would require changes of the storage implementation itself.
Julia asked abot time line for having the prototype 'sort of ready to use'. Adrian answered that the plan is for September.
Gavin told that CERN would be interested to send storage space accounting records for non-traditional storage, like for example for ElasticSearch. John told it should be possible. Julia mentioned that ES metrics on the UI should be disentangled from the traditional storage. John confirmed that it should be possible.
Next meeting will be most probably in August. The date will be communicated later.
There are minutes attached to this event.
Show them.
