1. Paper modeling : M. Abolins has had a discussion with V. Vercesi
and S. Tapprogge. The conclusion was that a two-way communication between
the modeling effort and the PESA group should be maintained, as results
of modeling have an impact on subjects to be studied by the PESA group
and vice versa.
After a discussion on how the 2.10**33 luminosity should be handled,
it was concluded to adopt for modeling the following approach : the baseline
is still the low luminosity menu, but with a reduced rate for validated
muons : this should be 5 kHz in stead of 9 kHz, i.e. the TRT and / or pixel
scan is done at 5 kHz in stead of at 9 kHz. For the 2.10*33 case it will
be assumed that all rates are multiplied by a factor of 2 (i.e. scan at
10 kHz) and that the processing time for the scan increases with a factor
of 2 (there is an almost linear dependency of the execution time on the
occupancy, see DAQ note 2000-031 ). The processing time for the scan at
1.10**33 will be taken from DAQ note 2000-031, and scaled appropriately
to take into account that the use of 4000 MIPS processors is assumed. The
execution times for the other algorithms have to be revisited again, S.
George suggested that the TP and back-up documents contain relevant information.
There was a short discussion concerning modification of other parameters.
For the merging speed 80 MByte/s has been used, K. Korcyl has measured
170 MByte/s on a 400 MHz Pentium machine, so assuming 160 MByte/s seems
to be justified. It seems reasonable to keep other parameters (e.g. the
task switching time of 10 microseconds) and processing models the same,
but this has not been firmly decided.
2. ROS modeling : R. Cranfield described the document on UML diagrams containing an example of how a model can be documented by means of UML diagrams. It is based on a small part of the DAQ-1 Dataflow software. J. Vermeulen showed the diagrams and argued that a good approach consists of starting with a class diagram. For each relevant class the next step is to make one or more activity diagram(s), proceed with collaboration / sequence diagrams (with "Together " one can alternate easily between the two different representations) for describing interactions between the objects and finally with the help of one or more state transition diagrams the code of the model can be constructed, as was illustrated with a small piece of C++ code to be used in the context of simdaq. R. Cranfield emphasised that the diagrams are not meant to document software on which the model is based, but to document the model itself. K. Korcyl asked about the choice of Together as tool and also asked why a tool like Together is better than a general-purpose drawing tool. J. Vermeulen answered to the latter question that Together maintains an administration of objects and methods defined which can be used in new diagrams, e.g. for labeling actions in a sequence diagram. It also takes into account the structure of a diagram when moving a part of a diagram. Together is recommended by the off-line group and available for academic institutions without cost for academic work. After a remark by J. Vermeulen that the use of the tool is intuitive it was decided that a small demonstration would be given by J. Vermeulen on the day following the meeting.
3. HLT modeling : S. Wheeler reported that there will be a discussion with H.P. Beck to define the boundary for modeling between the HLT part and the Data Collection part of the model, in particular with respect to the model of the DFM. The SFI and SFO are considered to belong to the HLT domain. The EF processors probably can be modeled by means of a simple delay between the arrival of event data and the production of the results. S. Wheeler will concentrate on the EF part of the HLT system, P. Golonka will look after the LVL2 part. S. Wheeler also reported that hardware from Alberta has arrived at CERN (21 2-processor machines) which will be used for EF studies and which will provide input for modeling.
4. Data Collection modeling : K. Korcyl and P. Golonka are developing a model for the TCP/IP stack. K. Korcyl concentrates on methods of measurement and parameterizing results of the measurements, P. Golonka develops code implementing different control methods available in the TCP/IP protocol (e.g. control of data retransmission in case of transmission errors). It is estimated that another two weeks are needed for completing the measurements.
5. Ptolemy issues : R. Cranfield reported that he is debugging
the multi-step processing and that this is working correctly if a single
step is specified.
K. Korcyl and J. Vermeulen had looked into the remaining discrepancies
between the Ptolemy and simdaq testbed models. It has been found that one
parameter in the simdaq model had not been used in the Ptolemy model. A
better agreement is now observed for the more complex setups, but for the
simple setups the discrepancy has become larger. K. Korcyl remarked that
with the same input parameters and the same models the same results should
be found. J. Vermeulen remarked that the difference in results may be due
to the relative priorities assigned to the different processes in the LVL2
processor.
6. Simdaq issues : J. Vermeulen had nothing to report.
7. Chiba city : R. Blair mentioned that the problem with the lock-up had been solved, but the next problem had appeared. This is caused by the two IP addresses available per machine (one for Fast Ethernet, one for Myrinet) and messages being sent using the wrong IP address. R. Blair will pass configuration information to the Rumanian group interested in modeling the system, a visit of group members to CERN is foreseen in the coming weeks.
8. Workplan : R. Blair mentioned that we had a different approach when specifying our workplan than the other groups had. This can be attributed to the fact that we do not have a direct representation in the Steering Group. He also said that there had been no response of the dataflow groups with respect to the workplan. K. Korcyl remarked that Data Collection task number 26 on architecture connected to the workplan and that R. Blair was one of the responsibles for this task , i.e. members of the different data flow groups and also working on modeling should feel responsible to check that the workplan of their group with respect to modeling matches with the modeling workplan. J. Vermeulen concluded from the present status that the work is progressing reasonably in line with the milestones specified in the workplan.
9. Next meetings : As agreed earlier, there will be a video meeting on March 13. The next meeting is to be held at CERN during the trigger/DAQ week in the first week of April. There was some discussion on the meeting frequency, one meeting per 3 weeks seems to be reasonable.
10. AOB : R. Blair mentioned that it was brought to his attention that the transmission of calibration data could need to be modeled. The question was asked whether calibration data would need to be transmitted during normal data taking. The asnwer is not clear, this needs to be looked at.
Notes by J.Vermeulen, 23 February 2001