System performance modelling WG meeting

→ Europe/Zurich
31/S-028 (CERN)

31/S-028

CERN

30
Show room on map
  • Local: Andrea Sciabà, Markus Schulz, Jan Iven
  • Remote: Alessandra Forti, Helge Meinhard, Johannes Elmsheuser, Daniele Bonacorsi, Andrea Sartirana, Michel Jouvin, Pepe Flix

Markus and Andrea show some slides about the follow-up of the WLCG/HSF workshop in Naples. Highlights from the discussion follow.

Johannes, about prmon: it could be enhanced to do some simple parsing of the application logs to extract e.g. information about the time boundaries of the different subtasks executed in a job.

Jan, about manpower estimation, says that there is a scaling factor of manpower with resources.

About testbeds: the existing Datalake testbed is VO-agnostic, we should profit from it (and other existing testbeds) and avoid doing something in parallel. Currently it is focused on storage, maybe it should be integrated with more CPU resources. Markus proposes the model of the former WLCG PPS (pre-production service).

Johannes will give at a future meeting a presentation on a study done in ATLAS about the impact of job failures and subsequent resubmissions.

About software efficiency: it is not obvious to draw a line between our WG and the software developers; we certainly can look at the behaviour of current code, e.g. measure inefficiencies due to low level parallelism.

On modeling of computer centres, Markus proposes to invite a student from NICA to give a talk. Jan is working on a graph-based model which could be interesting.

On the issue of calculating costs in terms of money, which is rather delicate, Markus suggests that no results should be made visible anywhere, but it is okay to have a spreadsheet that can be used to calculate costs (and publish it without numbers).

There are minutes attached to this event. Show them.