Sprecher:
Bartosz Sobol(Jagiellonian University), Dr.Grigory Kozlov(Goethe-Universität Frankfurt am Main), Sergei Zharko(GSI Helmholtzzentrum für Schwerionenforschung GmbH(GSI)), Sergey Gorbunov(GSI Helmholtzzentrum für Schwerionenforschung GmbH(GSI))
VTFeeder: feature complete.
KFPF: MR pending (MR 2377). In the current approach, the event selector(s) set a flag in the CbmEvent structure. The final selection then loops over the events and selects those with the proper flag.
Multithreaded QA: done. The needed plots have to be defined (collection in a Redmine ticket).
Tracking on CPU: STS-local and global tracking are available both with timeslice and with event input. Global tracking is about 50% slower, which is to be investigated. There is a 20% inefficiency when attaching TOF hits, which also needs to be understood. There is some minor problem with TRD and TOF hits (the hit finders do not deliver them time-sorted), which has lower priority.
Tracking on GPU: A MR will be finished otday. There is a problem with running STS ereconstruction on CPU and tracking on GPU. The STS hit finder allocates (too much) fixed memory on GPU; settings to tune the memory usage are required. The GPU tracker can be used on event input, but in this case no large acceleration is expected.
DDS: Progress on the SCA (close to ready). Tests were done locally; tests on VIRGO still need to be done. A fix for timers and monitors with XPU is already merged.
S. Gorbunov reports that the amount of STS hits is smaller by a factor of three compared to event-by-event simulation. This needs to be verified and understood / fixed with high urgency.
15:30
→
16:00
Benchmarks of the online binary30m
Sprecher:
Andreas Redelbach(Frankfurt Institute for Advanced Studies(FIAS))
The benchmark numbers still refer to timeslice processing, not to event-by-event reconstruction. Scaling with the number of threads is observed, but with a slope much smaller than unity.