1. Paper modeling : M. Abolins mentioned that it is desirable to have the possibility to choose between more than two menus in the spreadsheet. Also the possibility to scale the CPU and merging speed would be useful. J. Vermeulen showed an overview of FEX processing times found in recent DAQ notes, see this draft document. In particular for processing of the inner tracker data considerably longer execution times are found than have been assumed previously (the values assumed were "informed guesses" rather than obtained from measurements). J. Vermeulen also mentioned that results of the recent measurements on CPU utilization associated with data transmission via Gigabit Ethernet can be incorporated in the paper model. Missing in the paper model are the ROBs receiving data from the LVL1 RODs and the RoI builder RODs, while also passing of the LVL2 result data to the EF is not taking into account. The data streams involved are important for event building and have to be incorporated in the model in a later stage (it is probably justified to neglect these data streams in first order). A possible staging of the calorimeter RODs and changes in the lay-out of the detector also have consequences for the model, for the present round of paper modelling we do not take these into account. Later in the meeting (in the discussion on Simdaq) the issue of the 11.111 % added to nominal processing times (10 % of the total) was mentioned : there is no clear motivation for doing this, so this 11.111 % can be omitted.
2. ROS modeling : R. Cranfield reported on progress in producing an UML description of a model of the ROS. The diagrams produced so far are available here ("atdf" refers to ATLAS data-flow, the focus is on the present software, "ROC" refers to the Ptolemy model of B. Rensch), together with the Powerpoint files with slides produced for the meeting and the document with advice already produced earlier. A document on the ROS model is to be produced. It was remarked that it is not yet clear whether the approach is succesful, as an implementation starting from the diagrams has not yet been produced.
3. Data Collection modeling : K. Korcyl and P. Golonka have continued with the model for the TCP/IP stack. Measurement results obtained with the setup in bldg. 513 are used together with the OPNET environment for checking the parameterization. The next step consists of implementation of the model in at2sim in he form of "TCPBroker" objects, which are inserted between objects representing application software processes and objects representing network interfaces. K. Korcyl mentioned that the newest generation of switches show a behaviour which is different from the behaviour of switches previously used. New switches break up incoming frames in much smaller cells. Queueing of these cells for the same output port may result in much longer latencies than observed for older switches, as a frame can be output only if all cells corresponding to it have arrived in the output port. P. Golonka mentioned that he is making use of the "Linux trace toolkit", which makes time stamps inside the kernel, for profiling of the linux kernel. He has also found a potentially interesting "light weight" implementation of TCP, consisiting of 5 kByte of source code and which functions in user space. P. Golonka has started to use "Together" for producing a UML description of his models.
4. HLT modeling : S. Wheeler could not be present, but is still waiting for a new PC to arrive to start work on producing UML diagrams.
5. Reference testbed model : K. Korcyl and J. Vermeulen are working on a document (draft of 16 April) showing how the testbed results with the help of simple formulas can be described accurately for most of the results. It should also include new results from the Ptolemy and Simdaq models. At Rutherford Lab detailed measurements on the behaviour of the LVL2 reference software have been started, the results of which may help with understanding the remaining differences between results from measurement and computer models, see also the presentation by F. Wickens of 6 April.
6. Ptolemy : P. Golonka will take over the multi-step menu code, but has not had yet time to do so.
7. Simdaq : R. Scholte reported that he included in Simdaq support for choosing during run-time the sizes of the event fragments on the basis of probability density functions (re-using code for choosing processing times on the basis of probability density functions). Preliminary studies with distributions for the SCT, Pixels and TRT data for the low-luminosity trigger did show only minor changes in the distribution of the LVL2 decision time. J. Vermeulen mentioned that support for the Ethernet link protocol, necessary for the reference testbed model, has been included in Simdaq.
8. "Chiba City" : R. Blair reported that small configurations work, but larger ones have problems, help is needed to resolve the problems. He also remarked that the system has to be shared with other users, this has to be taken into account when doing measurements and when interpreting the results. Modelling of the system will be discussed with M. Brezuleanu, member of the Rumanian group interested in this activity.
9. Workplan and status in relation to workplan, presentations on 4 and 6 April : J. Vermeulen concluded from the present status that the work is progressing along the lines specified in the workplan, but some items (work on "Chiba City" and on interpretation of the testbed model measurement results) proved to require more effort than originally anticipated. The documented results of the "Chiba city" runs and the phase 1 model in Simdaq (this concerns implementation of the event building strategy as used in the DAQ-1 project) (both had milestones for March 2001) are late. Documented testbed results of computer models were due in February, a spreadsheet with results was available at the time, as reported a document is being written. The work on UML diagrams (preliminary diagrams due in May) and on paper modeling is progressing more or less as anticipated. The presentations on 4 and 6 April were discussed. On 4 April the emphasis would be on measurements required for developing component models, while on 6 April the presentation would be on the changes in the paper model parameters, modelling the TCP/IP protocol and the testbed models. K. Korcyl and J. Vermeulen would try to rerun the Ptolemy and Simdaq models in time for the presentation (note : after the presentation work on both computer models have been continued, resulting in improved agreement between paper and computer models for the configurations where the rate and latency could be calculated with a simple paper model). The work on describing models in UML would be presented also on 6 April as one of the presentations on the ROS.
10. Next meeting : The next meeting will be a video meeting at Thursday 26 April, 16.00 hr CERN time.
11. AOB : There were no further items to be discussed.
Notes by J.Vermeulen, 16 April 2001