System performance modelling WG meeting
- Local: Andrea Sciabà, Markus Schulz
- Remote: Graeme Stewart, Johannes Elmsheuser, Concezio Bozzi, Haykuhi, Costin Grigoras, Jan Iven
Andrea presented an update of his work on running prmon on our reference workloads. Markus proposes to add iowait to prmon.
We should also measure the impact of remote access on our workloads. Johannes will set up the input of the ATLAS derivation reference job so that it can remotely read it from EOS. Andrea will run it in this configuration.
Concezio points out that the LHCb example job is configured for one event, and this can be easily changed.
About our talk at the next pre-GDB, we will show only the activities relevant to the benchmarking comunity; also Servesh will give a talk, on Trident (his system profiling tool based on perf).
Markus reports that a student wrote a report comparing the cost of local resources to cloud resources for CERN. The numbers cannot be disclosed, but the categories considered for calculating the cost are interesting for us. He will share them with the WG.
Andrey Kiryanov pledged some resources to run remote access tests.
We still miss networking from our resource estimation models.