Monitoring Farm

[ Up ]

The Monitoring Farm (MF) works with events which have been accepted by the HLT, and have been hence sent to storage. Algorithms running on the MF subscribe to all or a certain trigger-type of events. The MF has no direct access to the DAQ contrary to the EFF.

Sub-detector monitoring and analysis tasks will be running on the MF. There is no event by event output from this farm: the only results are provided in term of statistical information, i.e. histograms.

Contrary to the EFF, each node in the MF will run different monitoring code(s), to make each monitoring task simple, allowing easier maintenance, and decreasing the sensitivity to crashes of someone else’s code. Each detector may have several somewhat independent monitoring tasks. We imagine a few nodes per detector for this raw data monitoring. This means that ~50 nodes should be largely enough to run the standard monitoring for the whole LHCb detector. In case of need, the same code could run on several nodes, provided each node gets different events.

The reconstruction (Brunel) will be run in the MF on a small sample of events. This will be done in parallel on several nodes, to get an appropriate rate of reconstructed events. To fix the idea, we are thinking of the order of 50 Hz. The result of the reconstruction will be packed in the Raw Event format, to be handled by yet another layer of Buffer Manager and made available to the relevant monitoring tasks on one or a few nodes.

The overall system is schematically represented here.