Michigan: M. Abolins
Argonne: R. Blair
UCL: R.Cranfield
Cracow: K. Korcyl (via telephone)
NIKHEF : J. Vermeulen
CERN: S.Wheeler, S. Gadomski, P.Golonka,
1. Paper modelling: New input for paper modelling has been received (in the days after the meeting more information was received). The spreadsheet has to be modified (done at the time of writing these notes) and the results have to be documented.
2. ROS modelling: Models are available, as reported in the previous meeting, but timing parameters are not readily available. The recent start of the redesign of the ROS software has diminished the motivation for running the models. R.Cranfield noted that implementing the models has been quite time-consuming, one can ask whether the model is too detailed. J. Vermeulen had at the ROS workshop posed the question whether we could obtain from the models new information. This in view of the test results obtained, which indicate that either the ROBIns used or the network I/O are the limiting the performance. J.Vermeulen suggested that an item of interest to study is the different behaviour observed for x % RoI request fraction and y % accept fraction vs. y % RoI request fraction and x % accept fraction. The difference probably is due to how passing of requests and data is organised in the software, while the values of the various processing times probably do not matter much.
3. Data Collection modelling and HLT modelling: P. Golonka showed diagrams for the L2PU . These diagrams are not really UML diagrams. This gave rise to a discussion on the merits of UML and of related tools. P. Golonka did not feel comfortable with UML and did not want to spend time on learning UML and a tool like Together. He has been using Dia, which is contained in Linux distributions. It is a drawing tool for the Gnome desktop environment, which also provides support for drawing of UML diagrams and is a less computer power requiring and a much simpler (but also less complete) tool than Together. R. Cranfield noted that UML is convenient and that we should use it. Concerning the implementation of the model described by P. Golonka: it is not clear whether the model for the LVL2 processor implemented by S. Wheeler can be re-used
K. Korcyl reported that there are still problems with the "strawman architecture" model. It takes 8 hours on a 1 GHz machine to simulate 1 s of LHC time. The first 6 hours of these 8 hours seem to be used for setting up, it is unclear what is going on then.
S. Wheeler showed diagrams from a document describing the testplan for the calibration of the DFM model . S. Gadomski will take care of doing the measurements.Ê
4. Next meeting: The next meeting is planned for Wednesday 27 March, 15.30 h CERN time. After some discussion (after the meeting via email) it was decided that this will be a telephone meeting.
J. Vermeulen, 1 March 2002