Integration of the new benchmark in the accounting workflow

Wednesday 25 Sept 2024, 11:30 → 12:00 Europe/Zurich

Zoom link:

https://cern.zoom.us/j/63307717078?pwd=OXFSeWJPUi90V0IxUHVrOFBhZVpMZz09

 

 

Attended:

Adrian, Alessandro, Carlos, Ivan, Maria, Julia

APEL

There were just few days since last meeting, no much progress reported by Adrian. 

Maria submitted a GGUS ticket since CERN data was not visible in the staging portal, looks like due to a wrong destination configured on the CERN side, should be fixed now.

At the next meeting Adrian will overview the status of site testing after dicussion with Greg.

Adrian will provide APEL status report presentation at the next WLCG Ops Coordination meeting, next Thursday.

Julia reminded Adrian about open issue of reporting from one of the US sites

Discussion regarding time to work conversion factor

People working on benchmarking are crosschecking benchmarking factors as they should be defined based on the benchmarking tests with the factors used for the accounting. Julia checked information we have on the accounting task force twiki and was wondering whether it was uptodate. Whethere GLUE1 attributes in BDII are still used by APEL? Alessandro told that for ARC old GLUE1 attributes are still used, while for HTCondor GLUE2 ones are used. Julia asked Alessandro and Adrian to provide fresh documentation regarding which attributes are used and how APEL then calculates wallclock and CPU work based on time. Julia moved docs on the twiki under Archive, not to confuse people, and created a new section for the latest docs (https://twiki.cern.ch/twiki/bin/view/LCG/DataNormalization#Data_Normalization_as_of_25_09_2)

Julia asked Adrian whether APEL keeps benchmarking conversion factors (some snapshot) in the APEL DB. Adrian told that conversion factors are contained in the job records and therefore are being kept for some time in APEL DB and  can be retrieved from there. Julia asked Adrian whether it would be possible to retrieve the list of conversion factors used for each cluster to provide the list to benchmarking people. Adrian will look into it.

For the normalized records, when calculation is done by the site, we do not know what every site is doing. For CERN Maria explained that each job reads a conversion factor from the WN and this one is used for conversion, there is no single value used for the cluster.

Julia asked Adrian whether conversion happens on the client or server side. Adrian will check.

Collaboration with QUANTA

Alessandro told that all necessary papers have been signed and sent. Ivan told that from the beginning of October he would try to allocate at least one developer to work on the project, and later starting next year up to 3.

Next meeting is Friaday the 4th of October 11:30 Geneva time.

 

11:30 → 11:50 Introduction

Speaker: Adrian Coveney (STFC UKRI)