System performance modelling WG meeting

→ Europe/Zurich
31/S-028 (CERN)

31/S-028

CERN

30
Show room on map
  • Local: Andrea Sciabà, Andrea Valassi, Markus Schulz, Jan Iven
  • Remote: Catherine Biscarat, Graeme Stewart, Renaud Vernet, Johannes Elmsheuser, Gareth Roy, Oxana Smirnova, Concezio Bozzi

Renaud presents the model he developed to predict the capacity of a computing centre as a function of the annual budget under certain assumptions. He is successfully using it for the IN2P3 Tier-1 centre. It works in a situation of steady state (that is, it is not applicable right at the start). The uncertainties are dominated by the tape system, whose cost evolution is much more complicated to estimate.

The costs for CPU and disk include the equipment (racks, switches, etc.), but manpower is not included. The tape costs include the tape library, its depreciation, its expansion and the evolution of cartridge costs per capacity.

Andrea V. comments that sometimes expenditures can be delayed at particular times, Markus that the network is usually upgraded at step increments, Jan that one can separate the part of the network costs which are proportional to the capacity (and add them to the CPU/storage costs) from the core network infrastructure costs. Renaud points out that many centres must spend their annual budget every year and cannot let save it for later.

Going to the review of the task list...

Graeme reports that prmon has been almost completely rewritten in a more factorised, plugin-friendly way to make the addition of new metrics much easier. Markus will put a student who will come in summer in contact with Graeme, as the project description is about improving the measurement of metrics.

Finally Markus presents some ideas on how to avoid duplication of work. Sometimes activities overlapping with ours are "discovered" by chance. We might have a specific task on the coordination of related work and ask people to publish their work plans. Andrea V. agrees but points out that we cannot stop other people from working; Markus proposes to still have some informal synchronisation. Graeme adds that also in the HSF they make efforts to avoid duplication of work.

Markus invites everybody to have a look at his Google slides (linked in the agenda) on ideas for a HL-LHC resource model and put comments on them, for a discussion at the next meeting.

 

 

 

There are minutes attached to this event. Show them.
    • 16:30 → 16:50
      Site cost modeling 20m
      Speaker: Renaud Vernet (Centre National de la Recherche Scientifique (FR))
    • 16:50 → 17:10
      First thought for a HL-LHC resource model 20m

      https://docs.google.com/presentation/d/1FrdgtgxrdF9sC5ffmmouJKh9sydfiwIn6I8-hvNTcZo/edit?usp=sharing

      Speaker: Markus Schulz (CERN)
    • 17:10 → 17:25
      Round table 15m
    • 17:25 → 17:30
      AOB 5m

      https://docs.google.com/presentation/d/1FWtofYvxgF7_mf-CWRNyMKBPdIEUeeTgHIJh_esnQ1c/edit?usp=sharing