Notes on the LVL2 video meeting of September 21 1998, 16.00 -17.25 CERN time

Particpating institutes : ANL, CERN, MSU, NBI, NIKHEF, RAL/UCL

1. paper model status / update

JV mentioned that for updating the paper model information is required on :

- Trigger menus : for modelling LVL1 trigger menus with exclusive rates for 
the various menu items are essential, LVL2 trigger menus are important for the 
output of the LVL2 system, for modelling the LVL2 system itself LVL1 menus are 
essential. The trigger performance TPR contains relevant information, but at 
the meeting it was not clear whether the information is sufficient. The 
document has to be studied, Nick Ellis said that he and Traudl Hansl can 
provide help when we can provide him with a wish list.

- ROBs : information is needed on the number of ROBs per subdetector and the 
mapping of the ROBS on the detector. Patrick le Du has informed JV that the 
ROD workshop in November will provide the required information. JV remarked 
that this information in January was said to be available in April. Peter 
Clarke has information on the TRT, Speranza Falciano on the muon detector, 
Rudy Bock will contact Speranza.

- Execution times of algorithms : the LVL2 and trigger performance reports 
contain relevant information. This has to be studied.

- Technology information : task switching times, network performance, etc. . A 
list of relevant parameters is required. Nick Ellis remarked that it is 
desirable to circulate a list 1-2 weeks in advance of the Chammonix workshop 
so that we can try to define the values of the parameters at the workshop. JV 
will take care of this.

After some discussion JV concluded that a new paper model has to wait until 
January. He also mentioned that, in view of our approximately 4-monthly 
meeting schedule a meeting on modelling, again combined with a ROB meeting, at 
that time is desirable. Patrick Le Du has suggested to organise this meeting 
again in Amsterdam. At this meeting then a new paper model should be existing.

2. Generic full model status / update

JV mentioned problems with agreed contributions to further development of 
simdaq-C++. The experience has been that a considerable investment in time is 
required, partially due to the required knowledge of many details of the 
program, which in particular is true for tasks like implementation of "Model C 
processors" and SFIs and of sequential processing. Alexis Amadon has not been 
able to contribute in the way as we had foreseen. Peter Clarke last week 
concluded in an email message and confirmed this in the meeting that UCL has 
come to the conclusion that its resources can be more efficiently used for 
other tasks than implementing support for sequential processing, as agreed on. 
(Not mentioned in the meeting was that also Kris Korcyl has been working on 
the program, but also had problems with making progress). On the other hand 
Bob Blair has looked to allocation strategies in the supervisor and could 
contribute there quickly, as only limited knowledge of the rest of the program 
was necessary.

JV mentioned that the current state of affairs is not catastrophic, as many of 
the open questions concerning implementation of the full model that existed at 
the time of the workshops in Argonne and Marseille now are answered. It is 
expected that this fall the full generic model can be completely implemented 
by JV without further external help. When that is done it is expected that it 
will be much easier for other persons to contribute e.g. with implementation 
of models of specific switching technologies etc.. The reason is that it is 
likely that issues that require a detailed understanding of all or of 
considerable fraction of the details the program have then been dealt with and 
do not require further attention.

The present status of simdaq-C++ was summarized :
- Improvements aimed at a reduction of the possibility of introducing bugs 
have been made, objects representing physics event information and objects 
representing messages flowing in system are now decoupled,
- A bug in the code for the RSIs was found and removed, the model of the 
DAQSwitch was improved (the model did only allow transfer of a message to an 
output port until that output port had finished the output of a previous 
message. This affects the shape of the model B latency curve as shown in the 
LVL2 TPR, but the general behaviour (curve with oscillations) is still the 
same,
- the model for the Model C processors and SFI objects as described in DAQ 
note 70 has been implemented,
- the results of DAQ note 81 (results of DS link based model B demonstrator) 
have been reproduced,
- sequential processing has been partly done, starting the weekend before last 
weekend after a message from Peter Clarke on problems with respect to further 
work in the coming months. An extended format of the menu file has been 
defined for specifying processing sequences, the code for reading in this 
information has been implemented and debugged, (physics) event generation 
based on the information in the menu file and using a modified representation 
of the physics event data has been implemented and debugged, the 
implementation of the processing itself is incomplete but is not expected to 
be problematic.

Rudy Bock mentioned that it could be of ineterest to use events also used for 
the reference testbed software as input, also making use of the decision model 
of the testbed software. JV said that this probably would be straighforward, 
but that one has to be careful with small event samples (where one has to use 
the same events over and over again). However, the correlations present in 
Monte-Carlo data between different RoIs of course are of interest, so this is 
something to be investigated later.

3. Generic models of testbeds

The results obtained for the system described in DAQ note 70 will be 
documented and be presented in Chamonix. 

Bob Blair tries to model the model C testbed. The "ModelC processor" objects 
now implemented should be of use. He also pointed out the desirability of 
having a form of cookbook with how-to-do recipes for a programmer working on 
the program and stated that he will attempt to produce this type of 
information.

Rudy Bock asked for a description of the models in the program. JV said that 
the process models are documented in DAQ note 70, but that models of the 
switches and network connections are based on those in the orginal (MODSIM) 
version of the program and have not been really documented. JV agrees that the 
models should be documented, but first likes to complete the generic model 
implementation.

4. Full models of testbeds

JV asked first Bertram Rensch for comments and explained that tools like 
Ptolemy and OptNet could lead to a rapidly obtaining knowledge about crucial 
features of models able to reproduce the results. Bertram is looking for a 
well-defined activity that can be done almost independently of other 
activities. After some discussion JV mentioned the possibility to have a look 
to the SCI testbed results, but it was concluded that this is probably is not 
worth the effort. During the discussion the work on modelling Gigabit Ethernet 
of Kris Korcyl with OptNet and of Richard Hughes-Jones with Ptolemy was 
mentioned and JV suggested that Bertram could try to work together with these 
persons. However, this is a problem in view of the desire to work almost 
independently of other activities. No conclusion was reached and time started 
to run out, leaving only 2 minutes to discuss the contribution to the workshop 
in Chamonix.

5. Plans for Chamonix.

JV mentioned that he has asked Patrick le Du for an hour. Possible 
presentations : overall status and workplan, progress with and results from 
simdaq-C++, work / results obtained with Ptolemy, work / results obtained with 
OptNet. With a time of 50 minutes allocated to these presentations we would 
have 10 minutes left for discussion. Nick Ellis remarked that it is possible 
to ask for a longer time (1 1/2 hour). 

The connection to NIKHEF was cut off abruptly at 17.25 CERN time.

Jos Vermeulen, September 21 1998