CBM Online Computing Meeting

B. Sobol

• Improvement of the timers in the online binary is finished - see MR 2509. The top-level processing steps are covered, but this is not the last step.
• Working on a unified system for timing and monitoring - work in progress, timescale some two weeks.

N. Greve

• The timeslice fowarder is integarted in flesnet. For the update of cbmroot, see MR 2417. Some dependences are not available in the standard apptainer, which is why the TS forwarder is by default not built in cbmroot. It is, however, built in the flesnet CI.
• The container used for the Data Challenges, comprising flesnet and cbmroot, is currently built manually on lustre, so not taken from the container registry.
• Testing the new RDMA interfaces of flesnet; looks good so far.
• Alternative "unfair" scheduler, allowing to overload selected nodes: work in progress.

G. Kozlov

• Work on the GPU track finder (global tracking, both event-by-event and time-based) is finished; a MR will be issued soon.
• The performance of the CPU tracker in the event-by-event reconstruction was tested. As expected, the speed is not satisfactory.

S. Gorbunov

• Working on switching vectorization for KFParticle and CA track finder from VC to std::experimental::simd. 

S. Schütt

• Benchmarking of online chain (CPU) finished. Results will be presented soon.

V. Friese

• Re-running beam transport with proper settings, avoiding downstream beam-pipe interactions.
• Trigger study with realistic input data: work in progress.

Next Data Challenge: 28 October. Concentrating on GPU workflow. To be confirmed by FIAS guys.