Argonne : R. Blair
CERN : P. Golonka, K. Korcyl
Irvine : A. Lankford (part-time)
Michigan : M. Abolins
NIKHEF : R. Scholte, J. Vermeulen
UCL : R. Cranfield
1. Paper modeling : M. Abolins has contacted S. Tapprogge and
will see him during the ATLAS week in view of the study on possible rate
reductions. J. Vermeulen mentioned that in the ROS meeting of one day earlier
it was reported that in the Steering Group meeting was reported that the
first physics data taking periond will be in the second half of 2006 with
a luminosity of 2 * 10^33, i.e. twice as much as foreseen for low luminosity
running. A. Lankford clarified that the average luminosity over this period
is expected to be 2 * 10^33, the integrated luminosity should amount to
10 inverse femtobarn. The increased luminosity does create a new situation,
in particular the critical B-physics trigger rate may need to be scaled.
During the ATLAS week it will be tried to obtain relevant information.
J. Vermeulen proposed to assume for the next round of modeling the
use of 4000 MIPS processors (i.e. 4 times as much as used previously -
which corresponds to the processing power of current commodity processors)
in view of Moore's law, which the last few years still seemed to hold.
If the validity of this "law" also holds for the coming few years this
type of processors would be available in the beginning of 2004, two years
in advance of data taking. For networking it was proposed to assume the
use of Gigabit Ethernet with an effective transfer speed of 30 MByte/s
per link (i.e. well below the hardware capacity of 100 MByte/s). The question
was raised whether Gigabit Ethernet will become a commodity product. J.
Vermeulen answered that the newest Apple desktop machines have a Gigabit
Ethernet interface. This was confirmed by M. Abolins who reported that
he uses such a machine. There was general agreement that the proposals
made are reasonable.
2. ROS modeling : R. Cranfield and J. Vermeulen are working on a document with a few UML diagrams (documenting the TRG task in the ROS software) and describing how a model can be built from the diagrams. The document is foreseen to be ready in the course of next week, its purpose is to serve as guide line for modeling done by the other groups.
3. Data Collection modeling : K. Korcyl, together with F. Saka and P. Golonka is setting up a parameterized model of the TCP/IP protocol. Measurements are done on various nodes and tests with FTP are performed in order to obtain information required for setting up the model. It is estimated that the development of a model requires about two weeks of further work.
4. HLT modeling : S. Wheeler, who could no be present in the meeting, discussed together with K. Korcyl and P. Golonka the day before the meeting work on HLT modeling. K. Korcyl reported on behalf of S. Wheeler that she will concentrate on the EF side, in particular on the SFIs and SFOs (for the EF processors probably very simple models suffice). P. Golonka will look after modeling the LVL2 processors and will also maintain contacts with the PESA group in view of algorithm execution times required for the models. J. Vermeulen asked to see this also as part of the work on paper modeling and to maintain also contacts wit M. Abolins and himself.
5. Ptolemy : R. Cranfield is continueing with the multi-step processing. K. Korcyl has no news on the testbed modelling, there are still discrepancies between the Ptolemy model and the measurement and simdaq model results. Slightly different input parameters have been used. However, the LVL2 processor models and model parameters are believed to be the same, but differences of 10 - 15 % have been seen for cases where the processors are utlized for 100 % and probably determine the maximum rate. This has to be looked into further.
6. Simdaq : J. Vermeulen reported that a bug in the GUI (in the "SRGP" package from Brown University) has been removed with the help of a patch found on a Japanese web site. The bug caused a crash on Unix systems with displays with a colour depth larger than 8 bit. Further he has found that a .cpp extension for files with C++ source can be used on linux, Windows and MacOS systems, in the new release of simdaq this will be used as standard extension. He also reported that for code browsing on linux systems "Source Navigator", available under the GNU license from Redhat (http://sources.redhat.com), is a suitable tool. The tool can be used together with the "Insight" debugger (also available from http://sources.redhat.com), which consists of gdb and a GUI for it. A problem with Source Navigator is that it uses absolute file pathes which may need to be modified when a project is moved to another machine. R. Blair commented that Source Navigator is a mature product. It was previously available from Cygnus as commercial product (Cygnus has been bought by Redhat). Codewarrior for Linux is also available, a new release is due in March. R. Blair has tried the current release, he did not find the integration between project manager and compiler and debugger satisfactory. The new release for Linux may be better in this respect. He has been using "Code Crusader" which in the past was free, but since the end of last year is available only commercially. R. Blair asked when a new release of simdaq can be expected. J. Vermeulen answered that he thinks that it will take about a month to produce a new release that also will include UML diagrams of the code and a Source Navigator project for the linux version. The code has not been modified recently (apart from code in the Unix version of the SRGP package, as discussed earlier), R. Blair has access to the latest version and it can also be supplied to others on request.
7. "Chiba City" : The T2RC software (responsible for control of the reference software) gives rise to a lockup, R. Blair hopes with the help of R. Hauser to find the bug responsible for this soon. K. Korcyl will invite a member of the Rumanian group interested in modelling the system to CERN to discuss the switch architecture of the system (in about 2 - 3 weeks).
8. Workplan and status in relation to workplan : R. Blair has distributed the task list to the leaders of the three dataflow groups. He did not receive comments until now (apart from a "thank you" from David Francis). The first few items on the list with milestones in the near future were looked at. The first one consists of documented testbed results of Ptolemy and simdaq, milestone February 2001. The spreadsheet distributed by K. Korcyl gives an overview of measurements and simulation results, the aim is to have a document ready on 1 March (see also the discussion on Ptolemy, agenda point 5), which is in line with the workplan. The next item concerns results on Chiba City, milestone March 2001. The work is under way, see the discussion on this subject. For the full system model implemented with rough components (i.e. those used for the testbed simulations) the milestone is April, this also seems to be OK. The last item discussed was the reconfiguration and/or review of models to conform with the phase 1 system with milestone in March. For the Ptolemy model this is done, thanks to the integration of G. Lehmann's Ptolemy code, for simdaq parts of the phase 1 event building model have still to be implemented but it seems possible to do this before April.
9. Next meeting : It was decided to have a face-to-face meeting at CERN during the ATLAS week on Wednesday afternoon (21 February) at 14.00 u. The location will be room 14-4-030.
10. AOB : J. Vermeulen mentioned that he prefers to have video
meetings in stead of telephone meetings. This preference was shared by
the participants. Unfortunately the Monday afternoon slot is occupied by
the Data Collection and HLT telephone conferences, although it is also
still reserved for video conferencing. J. Vermeulen has learned from H.P.
Beck that moving these meetings to another time (so that the video slot
can be used again) presents a problem. He asked whether it would be possible
to organise a video meeting on Tuesday afternoons. M. Abolins offered to
investigate this. The result is that on Tuesday March 13 a video meeting
(16.00 - 17.30 CERN time) will be held. This date was chosen as it is about
half way between the February ATLAS week and the April trigger/DAQ/DCS
week. R. Cranfield remarked that the AH232 protocol may be of interest
as it provides a quality as good as ISDN via an IP connection, while this
protocol is also supported by the switch used for multi-point video meetings.
Notes by J. Vermeulen, 9 February 2001