M. Abolins, J. Baines, R. Blair, P. Golonka, V. Gonzalez Millan, K. Korcyl, S. Wheeler, J. Valenta, J. Vermeulen
1. Paper modelling: An overview of the current knowledge on the parameter values is available from the modelling web page. In the last month new results on processing times were provided by J. Baines. Preliminary paper model results were presented on Monday 22 April (slides). From the presentation by Chris Bee that day it became clear that more complex trigger menus are used for the paper model than the simplified menus used for the LHCC presentations. For low luminosity a different event building rate (1.4 kHz in stead of the 2 kHz of the paper model) was used for the presentations. J. Vermeulen is reluctant to change the numbers in the paper model. J. Baines remarked that the processing times do not include time needed for unpacking raw detector data. New information may be available after the summer. Concerning the desirability of changing the menus and event building rate now S.Tapprogge and V.Vercesi have to be consulted. After some discussion it was concluded that the document with the paper model results will not be updated, but that the short "Overview of parameters ..." document, which also contains tables with results, will be updated from time to time. Older versions of this document should be kept. Furthermore the TDSG will be asked to specify the paper model results needed.
2. ROS modelling: R.Cranfield was not present, in view of his work on the ROS URD it seemed likely that there were no new developments.
3. Data Collection modelling and 4. HLT modelling:
K.Korcyl has understood the long queues in the switches observed with the model of the strawman architecture: these queues are inherent to the system modelled. The Gigabit Ethernet link model of R. Hughes-Jones was replaced by a simpler model. However, this led to an increase of a factor 7 in running time of the model (7 hr is needed per s simulated time for 4.5 kHz event rate). It is being investigated what the cause is. A new person in Krakow is becoming involved in the work. K. Korcyl also reported that the calibration of multi-layer models of switches by Asgeir Bjorlykke in conjunction with the measurements done in bdg. 513. is going on.
P. Golonka does not develop the "TCPBroker" further, as the model is very compute intensive and as using TCP/IP for data collection is of less interest. The model may be still of interest for modelling the communication between SFIs and EF processors.
S.Wheeler reported that S.Gadomski presented measurement results on the DFM the day before the meeting (slides) obtained with the MAGNI cluster (600 Mhz Pentium III, Fast Ethernet). Netlogger time stamps were obtained, while also the maximum rate at which can be run was determined. The Ptolemy model requires small changes. For the SFI a testplan similar to that for the DFM will be written.
P. Golonka mentioned that there are no new developments with respect to the L2PU model.
5. At2sim:Ê There were no new developments to report.K. Korcyl proposes to reorganise the repository and to discuss this inside the Ptolemy group.
6. Simdaq: Nothing to report.
7. "Chiba City": R.Blair is still intending to have a look at running the Data Collection software on the system, once the software is more stable..
8. Workplan and status in relation to workplan: In the last meeting
the question was raised whether the phase 2A system should be modelled.
K. Korcyl remarked that the system is very small and is aimed at demonstrating
functionality. Performance measurement results needed for modelling present
a problem. The view arrived at in the meeting was that it is probably better
to concentrate on performance testbed modelling and collecting the necessary
information. For the various components the following persons are responsible:
- S.Wheeler: DFM, SFI
- P.Golonka: L2PU
- K.Korcyl: Switch models, ROB emulators
- R.Blair: Supervisor
- R.Cranfield, J.Vermeulen: ROS
The status will be discussed on the basis of this list in the next
meeting. Concerning the present status: S.Wheeler will try to find a person
who will look after the model of the SFI. K. Korcyl mentioned that measurements
on an HP switch (which resembles the BATM T5) will be done in Krakow and
that a model of the ROS emulators has been produced by Asgeir Bjorlykke.
9. Presentation of modelling update: An update on the status of modelling is to be presented in the plenary meeting of Friady 26 April and will be presented by K.Korcyl with emphasis on the developments on the network and perfromance testbed models.
10. Next meeting: The next telephone meeting is planned for Thursday 6 June, 15.30 h CERN time.
11. AOB: none.
J. Vermeulen, 25 April 2002