WLCG Accounting Task Force meeting
Julia (chair), Miguel (minutes), Maria, Eddie, Alberto, Pepe, Antonio, Constin
- Update on data validation (slides)
Noted that EGI Accounting portal (batch/pilot data) and Dashboard (payload data) don’t always match, for example because of pilot overhead, or because some sites have scaling applied to wall clock directly (it would be useful if systems that do this would declare it).
Despite, data validation exercise is still useful.
For CMS EGI usage is 30-40% lower from raw and scale, needs further investigation.
Pilot overhead should be in 0-1% overhead range. Above that it should be investigated and understood.
Need to do one off study of end-start to check that systems that report scaled wall clock.
Need to be careful with estimated measurement.
Actual time is the only metric which has no estimation involved.
- PIC Accounting (slides)
CMS multi-core pilots running single-core payload seem to be a source for discrepancy between Portal and Dashboard data.This effect should decrease when CMS uses multicore payloads everywhere, currently only 4% of jobs. CMS also works on re-sizable jobs which should allow to minimize the draining time.
Currently there is a bug in the Dashboard, wall clock of multi-core payload is not multiplied by number of cores (4% of jobs)
So currently Dashboard number is lower that what is should be. The fix is being provided by the Dashboard team. After that CMS comparison should be redone
We need the raw numbers (time which is not scaled to also be reported).
MJF might be useful for payloads but are not essential (or necessary) for accounting at the batch system level or hardware level.
- CMS Pilots (slides)
-Next meeting is on the 14th of July, since Thursday in two weeks is WLCG operations coordination meeting.
-Julia asked members of the task force to evaluate the new portal, the results of this evaluation will be discussed at the next meeting