WLCG Accounting Task Force Meeting

Europe/Zurich
513/R-068 (CERN)

513/R-068

CERN

19
Show room on map

Attended:

Andrew, Alessandra, Domenico, John, Ivan, Gavin Olof, Adrien, Julia

Discussion regarding integration of the non-pledged resources into APEL.

John commented on the suggestion from Ivan to use "Infrastructure" attribute for  pledged vs non-pledged resources. There is currently a validation on the APEL side which allows only "Grid" vs "Local" values for this field. In case this attribute is used for other purposes (values), changes on teh APEL side would be required, since otherwise APEL loader would reject such reports.

John suggested rather to map non-pledged resource to a particular site, which might not be a normal site in the GocDB meaning.

Julia asked whether Yandex cloud resources are benchmarked already. Andrew told, that not yet and this could be a problem to use them for the first data flow prototype.

Discussion regarding implications on APEL in case new benchmark is introduced.

One of the main conclusions that any change which would require changes in APEL client might bring serious implications because of deployment campaign.

There was a discussion regarding use of teh conversion factor between old and new benchmarks for the transition period. Domenico, mentioned that there is no evidence that using of a single conversion factor would be good enough, since the dependency might not be liner. Moreover , conversion factor can depend on the CPU model, while APPEL currently does not take CPU model into consideration.

Discussion regarding CU concept presentation form Andrew

People liked CU concept for migration

Domenico mentioned that Andrew's presentation is in fact covers two different aspects. One is CU concept, another one is how to feed-propagate benchmarking information. Looks like external DB for benchmarking information can be a good solution.

How to find compromise between benchmark accuracy (covering all aspects of processing and scaling well) and time required to run. Fast benchmarks certainly have an advantage, since they can be easily run in boot time (preferable mode) and/or even inside the job. On the other hand it is difficult to create fast benchmark which would be accurate enough.

AOB discussion

John asked about status of space accounting. Julia told that work in this direction started. There are two main directions of work. One is to work with storage providers on enabling storage space accounting reporting. And this will be done in collaboration with the WLCG DM steering group. Another one is to enable data flow from the source to the repository and developing UI. This would be done in collaboration of the CERN Monitoring team. Technical student will work on the second topic for  a year.

Action list

Julia asked Andrew to present his ideas regarding CU concept and benchmarking proposal at the April GDB. Andrew agreed.

Julia also asked John and Adrian to talk about possible implications on APEL in case a new benchmark is introduced at the April GDB.

 

Next meeting is Thursday 20th of April. The main topic is Storage Space accounting.

 

 

 

 

 

There are minutes attached to this event. Show them.