1. Paper modelling: In the plenary meeting of Monday 12 November it has become clear that a number of input parameters for the paper model has to be revised. Among these are the event building rate, LVL1 RoI rates, reduction factors for sequential processing, data volumes from the calorimeter and inner tracker, time needed for pixel scan. It was discussed how a new baseline should be established. The conclusion was not to wait for an initiative from the Steering Group but to try together with the proponents of the new numbers, N. Ellis, J. Baines, M. Levine, V. Vercesi, and, for technology dependent numbers also with R. Dobinson and K. Korcyl, to define a new baseline. According to the existing workplan an update of the current paper model should be available in March 2002. V. Vercesi was not present at the time of the discussion, but arrived later, and then commented that he would propose action from the side of the Steering Group. The paper model spreadsheet was used in the meeting for determining the effect scaling the B-physics trigger rate when scaling tthe LVL1 rate to 75 kHz. It was found that with the current assumptions and for the B-physics trigger rate scaled for the intermediate luminosity scenario (current LVL1 rate 80 kHz) at minimum 741 in stead of 506 processors in the LVL2 farm would be needed. The bandwidth required for the LVL2 traffic would increase from 2 to 3.5 GByte/s.
2. ROS modelling: R. Cranfield reported on progress with the UML description of the ROS software (see also his web pages). Almost all text and diagrams in the FrameMaker document on the UML description is stored in the Together project, but the document cannot be completely produced automatically: cutting and pasting is still necessary. The question was raised whether using UML diagrams is a good way to get to a model. The answer is yes, producing Ptolemy models from the UML diagrams is easy. Another point in favour of the UML diagrams is that they also provide detailed documentation of the ROS software, which is not available in another form. Once the models are available the values of many parameters need to be specified. It was suggested to determine from the measurement results which parameters are most important, to extract the values of these from existing or, if necessary, from new measurements, and set the values of the remaining parameters to zero. R. Cranfield, in his role of modelling group representative, also reported on work going on in the Connect Forum. The design rules and design documentation guidelines seem to be useful and it is interesting to see how well the UML documentation matches with these guidelines.
3. Data Collection modelling: Together is now available in Cracow. In about one month time UML diagrams of the high-level model of the TCP/IP protocol of P. Golonka as well as UML diagrams of the switch models of K. Korcyl are expected to be available. The next step would be the production of diagrams for the ROB emulators to be used in the testbeds. In collaboration with the bdg. 513 group broadcasts are studied and is work going on with respect to calibration of switches. A model of the "strawman architecture", with a hierarchy of swiches, is to be set up. The Data Collection software is installed in Cracow, 2 dual Pentium III PCs with a Gigabit Ethernet connection are available for calibration measurements. S. Wheeler reported that it is now time to start looking at the code for the DFM and the SFI and to see whether existing models are in agreement.
4. HLT modeling: S. Wheeler has received her new PC! There were no further developments to report.
5. Ptolemy: The Norwegian student, mentioned in the last meeting, is going to work in the group for R. Dobinson, work on modelling should be in the context of that group. P. Golonka will be back at CERN in December.
6. Simdaq: J. Vermeulen reported agreement between Simdaq and paper model results for the different scenarios studied with the paper model and showed slides with distributions on the LVL2 decision time and for the percentage difference for a number of averages computed with both models.
7. "Chiba City": Linux kernel 2.4 is now used, therfore the reference software had to be rebuild. The decrease of the rate for large configurations is not understood. It has been found that the problem also occurs for small configurations with a high maximum number of outstanding requests in the supervisor. It seems therefore plausible that message queues are overflowing. Taking into account the problems encountered and the fact that the reference software is more or less obsolete, R. Blair considers further work on this system as a "hobby". K. Korcyl will inform the Romanian group interested in modelling the system about the current situation.
8. Workplan and status in relation to workplan: On the basis of the discussion in the last meeting R. Blair will update the workplan.
9. Presentations: The presentations of R. Cranfield (slides) and K. Korcyl (slides) for the Wednesday and Thursday plenary sessions respectively were discussed.
10. Next meeting: 11 December 16.00 h CERN time via VRVS, K. Korcyl will book the virtual room.
J. Vermeulen, 22 November 2001