Choose timezone
Your profile timezone:
System performance modelling WG meeting
→
Europe/Zurich
600/R-001 (CERN)
- Local: Andrea Sciabà, Andrea Valassi, Markus Schulz, Domenico Giordano, Alessandro Di Girolamo
- Remote: Andrea Sartirana, Catherine Biscarat, Gareth Roy, Johannes Elmsheuser, Renaud Vernet.
Domenico: very good progress, we have (GEN)SIM workloads for all experiments working fine but for some connections to outside network.
Container building is automatic and we tested the conversion to Singularity images.
One way is to use the Docker image in Singularity, another way is to convert it into a Singularity image.
We are in contact with SFT to have a ROOT based analysis in containers.
LHCb started validation studies. We still need to containerise the other workloads
We need the experiment contacts to help on extracting the score metrics.
Frontier access must be rendered unnecessary.
Johannes: we'll try to find somebody to help you.
Alesssandro: you need a bit of time from many people. about containerisation, we need a named
person, maybe Lorenzo?
Andrea V.: for Alessandra the container is the goal, for us it's the means.
We need to have some help in removing network access, that's all.
Andrea V. gives is talk on testing the LHCb SIM workload with Docker.
Discussed the agenda workshop:
Markus: we need collect in a more structured way projects for students.
We could think of putting in a discussion on the relationship with DOMA (it could also be in the DOMA session).
Is it enough to have people in both activities? or better more formal cross reports?
Andrea Sc.: better to have reports for those of us who are not in DOMA.
Round table:
Andrea Sc.: I am analysing CMS popularity data for cache simulation.
Markus: the student (Corentin) picked up work again, studying latency impact depending on where data and the cache are located. He also picked up the latest version of the cms workload.
Markus: I am also looking at the life cycle of atlas data (replications, deletions, etc.).
I splitted TCO per unit of hs06 and tb/y by diff contributing factors and made the spreadsheet more readable.
Andrea Sc. also started collecting cost evolution data to check if 20%/y is still realistic (spoiler: it is not).
Gareth: we are seeing that cost is drastically underestimated; numbers are now preliminary and confidential.
Andrea Sa.: I sent some data to Renaud.
Catherine: I prepared the budget for France.
Renaud: Gareth, can you provide us with the ratio between what you replace and what you buy as growth?
Gareth: I will ask Dave.
Renaud: tried to contact sites, for estimates about predictions for tape systems as followup from my talk at archival WG.
I asked numbers about wan connectivity but never got an answer.
I looked at markus' spreadsheet, it looks like manpower contrib is quite low: our director found that about 50% of cost is on mapower. I have no more details, but for next meeting i'll try to have an official statement
and more data.
Markus: I suggest to copy the sheet and play with the input numbers. The manpower numbers for hours are what we really see at CERN.
We invested over the years in this installation, to have it as efficient as possible.
I didn't put the CERN paychecks, only hours per server/year. The number very dependent on local realities.
Catherine: also for the French Tier-2's we do see 50% for salaries.
Markus: maybe we should discuss in more detail what goes into that effort. For example, I only took people involved in installing and looking after the stuff and people rnning the batch, etc.
There are minutes attached to this event.
Show them.