Choose timezone
Your profile timezone:
HEPiX Benchmarking Working Group
Minutes:
- Manfred Alef: HEP Workloads as a Benchmark: Benchmark Validation (v. 2019-07)
- This talk is a follow-up of the presentation 2019-05-10, now including also ATLAS sim+reco, CMS digi+reco, and KV. The status of ALICE gen-sim and ATLAS gen is still unchanged. A very huge spread has been found in both workloads. The status of LHCb-gen-sim is also unchanged, it had already passed the tests (robustness, spread, and runtime).
- ATLAS sim is also robust, and its spread is about 4%. However, the runtime of around 45 minutes (fully saturated host) is very long. ATLAS reco is robust, but the spread is up to 16%, with a runtime of ~30 minutes.
- The CMS workloads are robust, spread is up to ~3% (digi) or ~2% (reco), and the runtime is not longer than 15 minutes per workload. Andrea Valessi mentioned that the input file for digi contains only 100 events, so each thread of CMS digi is processing not more than 25 events.
- KV is a well-tested, robust application with a spread of not more than ~5% and a runtime of ~6 minutes.
- AOB:
- Andrea V. mentions that he has found some issues with CMS digi, in contrast to the outcome of Manfred's talk.
- Both the reptest script by Chris as well as the several workloads have a '-n' flag. Unfortunately there is no unique interpretation of this flag. In the workloads it sets the number of simultaneously benchmark copies, and in reptest it sets the number of consecutive benchmark runs. To make it more clear Chris and Andrea V. will rename the flag ('-r RUNS', '-c COPIES').
- During the summer holidays we will schedule at least mid-cycle Sprint meetings.
There are minutes attached to this event.
Show them.