CERN: M. Abolins, R.Blair, J.Vermeulen
UCL: R.Cranfield
Cracow: K.Korcyl, P.Golonka
1. Paper modelling: New developments: the 5 kHz "other" trigger item of the LVL1 trigger menus is not taken into account in the paper model, the B-physics trigger started with validation of a MU6 or MU8 and consisting of a full scan or a mini-scan of the TRT and/or the SCT and Pixels is no longer seen as defendable, the validity of the standard 10 microseconds per I/O action was debated in other meetings in the ATLAS week and needs to be revisited.Modelling the "sliced" scenario is put on hold, as it is not clear whether it will be used
2. ROS modelling: R. Cranfield is waiting for the new version of at2sim.
3. Data Collection modelling, HLT modelling, at2sim: P.Golonka and K.Korcyl: message passing: preliminary work on interrupt coalescence has been done, still some fine-tuning is needed; ROS emulator (ROSE): work to be done on modelling a small test setup; Alteon ROS emulator: working on model; switch models: problems with crashing, to be fixed soon; L2SV: calibration data are collected; L2PU: end of month: better model, followed by implementation of the model; DFM: not yet completely working, model of tester setup at the end of the week; SFI: end of the month. Conclusion: at the end of the month all models should be available. Scripts describing various setups, including the baseline architecture also will be produced on a short time scale. R.Cranfield looks after LVL2 specific problems (sequential processing and ROB mapping, S. Wheeler can provide information concerning her work in this field). In the Data-Flow meeting earlier on the day it was concluded that it is preferrable to give prioirity to getting results for event building: event building is more demanding with respect to the networking than LVL2 processing, while the mapping of the detector onto the ROBs and the details of the LVL2 processing, aspects that could be problematic, do not play a role for event building.
4. Simdaq: J.Vermeulen has not yet finished the implementation of SFIs in the program (event building is done in the current model in the processors of the Event Filter farm).
5. Chiba City: The kernel buffer space may be a critical issue as may be the socket buffer size. Only if it is possible for a user without special privileges to set these buffer sizes at the values required further work with the Chiba City system makes sense.
6. Workplan: No further discussion.
7. Next meeting: Wednesday 5 March 15.00 h CERN time.
J. Vermeulen, 18 March 2003