The session started with an introduction by J. Vermeulen. He pointed
out that modeling is seen as a task within the dataflow group, not as an
independent activity. He asked Andy Lankford, who was present, to
comment. Andy indicated that while it was likely that there would
be a few people who would concentrate on modeling as a primary activity
most of the effort should come from the individual groups HLT, Data collection
& ROS.
P. Clarke expressed some concern that while this was desireable it may not happen. He pointed out that the Ptolemy work had partly been started to make this easier, but no army of people showed up to help. M. Dobson pointed out that there should be good involvement since modeling flags problems that these groups need to address.
R. Blair expressed the view that at the very least there needed to be
a laisson from each group that would take an interest and facilitate communication.
R. Dobinson worried that the work list did not include understanding the
test measurements. This, it was explained, was implicit in developing
"detailed
models". Several people indicated that "what we need" from modeling
has to be better defined.
A number of people indicated that UML descriptions would help in making system component models. R. Blair pointed out that this was on the list of deliverables from subsystems and would be mentioned in his presentation on the workplan.
J. Vermeulen showed some of the areas where work was needed (slides). Updating model data is needed and this would benefit from some work reported on by F. Wickens in an earlier session where fragment sizes, ROB mappings, etc. were being collected. Beyond this he indicated that several other items still needed to be updated in the generic model (implemented in simdaq) : fragment size distributions, as well as for B physics distributions number of electron and muon seeds produced by the track search, the EB model, adding QoS to the network and documentation with UML.
For detailed models the first step should be documenting the existing systems with UML. A number of people pointed out several aspects of UML: it is incomplete for modeling (this is understood and requires addition information to be added - like timing) and it ought to be part of the requirements/design process prior to starting any code. R. Hughes-Jones asked if there were tutorials planned to get people up to speed with UML. It was pointed out that there are web based tutorials available.
R. Blair next presented a preliminary workplan for the period up to up to the TDR, see his slides.
For the second item (measurements and modeling of the Argonne "Chiba city" system) R. Dobinson would like to receive details of the network used, to see whether the networking could be a limiting factor. Scalability of the TCP/IP suite is something that can be studied with the setup, as well as transmission of LVL2 and EB traffic via the same network.
The milestone for 1 May 2000 (availability of preliminary UML diagrams) led to a further discussion on the use of UML diagrams, in which J. Schlereth noted that it is desirable to have templates for these diagrams.
The milestone on "Evaluation of the modeling tools and selection of a primary and of a secondary tool" gave rise to a lot of discussion. Different views with respect to desirability and, if seen as desirable, with respect to the time when it should be done, were expressed. It was remarked that it may be easier to attract new people to work on modeling if a single tool is used, that the tools may be complementary, that there is virtue in having two different tools and that the UML diagrams could be sufficient to implement a model in each of the two tools. Also the desirability of UML diagrams was debated. The general feeling seemed to be that it has to be seen how well this works, the ROS system may provide a test case. R. Blair finally concluded that there is no conclusion on the evaluation and selection of the modeling tools and showed the remaining milestones on the workplan presented. The persons involved in the work are foreseen to be the persons working on modeling previously.
It is planned to organise a virtual room meeting (or telephone conference
in case of problems) each 3rd Tuesday of the month at 16.00 h CERN time.
The first meeting is planned for 28 November. The pilot project modeling
web page is to be used as web page for the time being.
J.Vermeulen showed a slide on new insights on the paper model (slide).
M. Abolins is going to look into the work items identified in the June
meeting for which P. le Du and J. Bystricky were responsible.
R.Scholte showed a new result on the event fragment size distribution for the SCT for low luminosity (slides). The average size is higher than assumed in the paper model and the distribution has tail that can be sdescribed with an exponential function.
R. Cranfield discussed the implementation of sequential processing in at2sim (slides)
K. Korcyl presented the status with respect to network and network component models (slides).
M. Dobson described his work on the integration of the Ptolemy LVL2 model and EB model from G. Lehmann (slides). A note is in preparation.
J. Vermeulen discussed two new developments with simdaq (slides) : a detailed model of the NIKHEF ROBIn hardware developed by R. Slopsema and implementation of the Ptolemy model of the pilot project Ethernet based testbed in simdaq. For the testbed model it was found that some tuning of the parameters was necessary to obtain reasonable agreement (at the 10 % level) between all measurement results and model results.
Notes by R. Blair and J. Vermeulen