● VT26-2 coordination: status, plans, missing features

Instructions for testing

We need to distribute instruction how to run cbmreco in order to enlarge the circle of people testing and benchmarking it. Since for the time being, this is only for the online group, the instruction should be put into our Redmine Wiki. S. Zharko and A. Redelbach are asked to provide this information.

Benchmarking

The runtime measurements of cbmreco should be broken down on several levels.

• Top level: Unpacking, triggering, event building, event reconstruction and selection, write output
• Second level: (Event reconstruction): STS reco, TOF reco, CA tracking, PV finding, KFPF analysis
• Third level: Breaking down of algorithms, e.g. STS reco.

The timers (XPUTimers) should be checked such that they cover all processing steps and are consistent between levels.

B. Sobol will have a look until next week's meeting.

The input for the DC in August contained a wrong beam pipe model, leading to very high background from interactions of the beam with the downstream beam pipe (see Redmine Issue 4157). This also influences our runtime mesaurements. V. Friese will prepare some input without beam background in order to test cbmreco with more realistic input data.

Runtime optimization

• For the CA track finder, an acceleration of about 10% was obtained by using larger SIMD vectors (512 bit are supported by the newer VIRGO machines). Additional acceleration os expected when using this also for the downstream tracking (track extrapolation).
• It was seen that the memory allocation of the STS hit finder took about 30% of the toal reconstrution time. This will be fixed (S. Gorbunov).

T0 determination

We will not have an algorithmic implementation of the T0 reconstruction from TOF data (see Redmine Issue 4161). For VT26, we will transport MC truth into the data stream through the BMON digis. This will be adjusted such that T0 signals are only generated for collision events, not for beam events (V. Friese).

MC truth

We will not have linking to MC truth in October. Studies of efficiency, e.g., for Omega reconstruction) must be performed by comparing to event-by-event digitization, probably with the legacy reconstruction scheme.

Beam model

The current beam model is constant intensity (Poisson process: the event time separation follows and exponential distribution). A macroscopic beam model, including spill pauses and intensity variations during the spill, will be implemented at a later stage.

TS Forwarder

Next steps are to included more metrices in Influx/Grafana and to adapt to the new flesnet interface.

Merge Requests

S. Zharko points to two Merge Requests which should be attended to urgently:

• MR 2328 (event builder)
• MR 2473 (process graph for time-based reconstruction)