Markus invites people organising meetings for individual task activities to advertise them and make notes available for the others.
Graeme is working to make the ATLAS Memory Monitor independent from Athena. Markus suggests him to contact David Smith and Servesh, as they are working on something similar.
Gareth illustrates his and Graeme's work to streamline the metric tables and add some practical recipes to extract the metrics. Lots of information are available from /proc/<pid>. He suggests a way to isolate a Linux namespace to collect metrics in a more directed way.
Markus suggests to have a dedicated meeting to discuss the metrics. In his own work on metrics, his goal is to take one workflow and try to fill the table. This can be done after the workshop in a sort of "hackaton". When this is done, it will be easy to turn these measurements into practical ways to create time series for metrics.
Markus will prepare a Doodle to organise such meeting, preferrably before the workshop.
Catherine did the exercise using updated numbers, separating the cost of resources (CPUs and disk) from the cost of the basic infrastructure, which adds 18-20% to the cost of resources. She is very confident on the accuracy of her estimates. The only point which is not clear is the cost of electricity, as she got different numbers from different people, differing by up to one order of magnitude! For example, for disk, half of the cost comes from electricity, which looks strange. She confirmed that the numbers come from the guy who "pays". She also pointed out that there is a difference between the price actually paid for resources and the vendor price.
Gareth commented that the numbers for resource costs match very well with the ones for the Glasglow Tier-2 (which incidentally does not have to pay for electricity...).
Jan asked to clarify the level of confidentiality of these numbers: it is agreed that they should stay inside our group. Gareth suggested to factor in inefficiencies in CPU usage; Markus mentioned a recent talk by somebody from the Beijing Tier-2 illustrating how they used backfilling to reduce the cost of inefficiencies.
Catherine suggested to consider also the cost for licenses. Manpower is a much more sensitive point, for that we can use man-years (or FTEs). The WLCG survey of a few years ago can still be considered an authoritative source.
We have to contact Frank to see if he can give the introduction talk; in case he is not available, Pepe is willing to give it [update: Frank is unavailable].
Gareth can give from remote the talk on workflow properties; otherwise, Markus can give it and Gareth can connect to participate to the discussion (this might be the most practical solution).
Catherine will give the morning talk on cost estimates and Renaud will drive the discussion and the "hands-on" in the afternoon.
Note-takers will include Andrea Sc., Andrea V. and Catherine (at least).
Andrea V. can give the talk on tools and techniques.
David L. accepted to chair the dicussion on the Run3/4 evolution.
It is agreed to have a meeting among the session leaders before the workshop.