Notes meeting on computer modelling, 14 January 1997, NIKHEF, Amsterdam

A. Amadon, Saclay, amadon@hep.saclay.cea.fr
P. Le Du, Saclay, ledu@hep.saclay.cea.fr
J. Bystricki, Saclay, bystri@ hep.saclay.cea.fr
R. Hubbard, Saclay, hubbard@hep.saclay.cea.fr
R. Cranfield, UCL, rc@hep.ucl.ac.uk
G. Crone, UCL, gjc@hep.ucl.ac.uk
P. Clarke, UCL, clarke@hep.ucl.ac.uk
P. Sherwood, Peter.Sherwood@cern.ch
R. Blair, ANL, reb@hep.anl.gov
J. Vermeulen, NIKHEF, i73@nikhef.nl
 

Due to heavy fog in Krakow, Kris Korcyl, who intended to come, has not been able to reach Amsterdam.

The meeting started with an overview of the current status. There are now several persons studying or working on Simdaq-C++ :

• Kris Korcyl has implemented support for sequential processing and is currently testing his code,
• Bob Blair has developed code for writing a file with a dump of the histogram contents in a format that can be used as input for PAW,
• Peter Clarke is starting to look into B-physics (low-luminosity) trigger menus and implementation of the TRT scan and missing energy trigger (which is similar to the TRT scan),
• Alexis Amadon and Jiri Bystricki are starting to study the program,
• JV has been working on an updated version of the documentation. A preliminary version was distributed to most of the participants just before Christmas and in the first week of January.

• How to organise and partition the work,
• The question of the desirability and the feasibility of having a second tool for checking the results of the current program. It was noted that cross-checking with measurement results of demonstrators and emulators also should be done.
• Which questions are to be answered. It was noted that generic modelling as well as technology studies need to be done,
• Definition of tasks and milestones,
• New insights in trigger strategies, number of ROBs per subdetector and mapping of the detector on the ROBs. It is expected that there will be an update on the strategies in March (i.e. new trigger menus and rejection factors for sequential processing) and a revision in the number of ROBs and in the ROB mapping in April.
• The paper for the IEEE RT97 conference should also become available as ATLAS DAQ note.

The second point of the agenda consisted of a tutorial presented by JV. This was based on an abstracted version of the preliminary and partial documentation of Simdaq-C++ 3.0 (available from the ftp directory with Simdaq-C++ 3.0). The most important technical details of the program have been dealt with and a number of technical questions were raised and discussed.

After the tutorial the next issues were dealt with :

• The question of the desirability and the feasibility of producing a second tool for cross-checking was discussed. The general opinion was that, although this may be desirable, it seems to be unrealistic to expect that we can produce such a tool for simulation of the full system with the level of detail now already present in Simdaq-C++, also when starting with an existing simulation package or language. However, the work on Ptolemy by NBI will be looked at, in particular by Peter Clarke. It was also remarked that for data analysis usually there is no separate development of different analysis programs for cross-checking purposes. Furthermore the paper model results also provide an important cross-check, while also many consistency checks can be made (and are made) internally in the program during running, as well as on the results.
• The organization of code management. It was decided to work for the time being without central repository, as this implies also the use of a common tool. It was noted that individual code management is done with RCS or CVS or in a primitive way (by storing snapshots of the code just before drastic changes are made). JV will try to act as librarian and to produce new releases on the basis of code received. In a later stage we may have to organise this better, but at present this seems to be a reasonable strategy.
• Assignment of tasks. As noted in the overview at the beginning of the meeting a certain distribution of the work had been agreed on earlier. This was discussed again and new tasks have been added, giving the following list :
• Peter Clarke + other members of UCL group : low luminosity menu's, support for TRT scan and for missing energy trigge,
• Kris Korcyl : support for sequential processing,
• Saclay : execution of all feature extraction and global processing for a single event in a single processor,
• Bob Blair : modelling of the supervisor,
• JV : implementation of pull model and of the RSIs.
• The next meeting. The plan is to have a next meeting in about 3 months time. It was proposed to organise a meeting in conjunction again with a meeting on the ROBs as many of the potential participants also have an interest in the work on the ROBs. A possible date and place is the beginning of May at UCL in London.
• Definition of what we are aiming at. There was no real discussion on this point, but it was noted that for the TPR this has to be written down, and that JV had to provide a note on what will be written. It was decided that a draft version would be circulated before submitting (Note : this has been done, the version submitted is appended).

Jos Vermeulen, January 28, 1998

Outline contribution on computer modelling, section 4.6 of TPR

Assumption : about 1 1/2 page available (for paper modelling and emulation each also 1 1/2 page, as the total length of the part on modelling and emulation is 5 pages)

1. Results aimed for

Keywords, phrases : dynamic model of system, study of latency and of buffer occupancies, cross-check with paper model, localization of potential problem  reas in system architectures of interest and when applying of technologies of interest, determination of key parameters with respect to the required system resources, definition of the best trigger strategy (parallel/sequential) for system architectures of interest.

2. Method

Keywords, phrases : discrete event simulation, SIMDAQ-C++, extendible program, high-level configuration management, multi-platform implementation, on-line monitoring of running program.

3. Present status

Keywords, phrases : efficient simulation of full model B architecture as used for paper model, reproduction of paper model results, work force expanded since last summer and does consist now of core group with persons from Argonne, Krakow, NIKHEF, Saclay and UCL.

4. Some results

Latency curves shown in Beaune and Marseille and in the plenary ATLAS meeting of December + new picture with results for two-layer switch. Explanation.

5. Outlook

First : completion of implementation and study of models used also for paper model, including study of the impact of the different trigger strategies (parallel/sequential) on behaviour of the system modelled, next : "generic" simulation (with relatively simple models) of laboratory setups and identification of main factors determining behaviour, next : detailed simulation of technologies calibrated using results of laboratory measurements and re-using code / models developed earlier for ATM, SCI and DS link technologies.