System performance modelling WG meeting
- Local: Andrea Sciabà, Markus Schulz, Evangelia Santorinaiou
- Remote: Gareth Smith, Michele Michelotto, Martin Gasthuber, Paul Millar, Concezio Bozzi, Johannes Elmsheuser, Oxana Smirnova, Alessandro Di Girolamo
Paul gives an overview of the DOMA QoS sub-working group. One of the point he raises is the need to have a concept of "cost" for the storage, which could be defined with the help of the cost model WG.
Various comments were made: Johannes makes the point that SSDs have little use for ATLAS (now at least) as input data is always pre-copied to the WN; which implies that also storage for WNs should be considered.
Following a question from Gareth, Paul adds that disk caches are not being covered, only "persistent" storage. The quality of service refers to the "promise" to deliver files in a certain way (durability, performance, time to deliver the first byte, etc.).
Alessandro points out that CRIC will contain a service description for storage resources and their capabilities. Paul comments that the labels he described in his talk to classify storage could be mapped to different categories by different experiments and not all of them might be relevant for them, or only for certain cases.
We then moved to the round table and the task review.
Concezio is working on providing an updated LHCb workoad, as the old one had problems with CVMFS. LHCb is working on a (very) large container that does not require access to CVMFS to run; it is a snapshot of their main workload and includes simulation, digitisation, reconstruction, trigger simulation and stripping.
Markus announced that the student who studied the dependency of throughput from memory and network latency will return and turn his code into a standalone tool starting from next month. Then he will work on parameterising metric time series.
Concezio mentioned that LHCB is working on the computing model TDR, and this work will be used for the resource needs estimation.