Minutes of T2UK mtg at UCL (from Erkcan)

Minutes of last face-to-face meeting

No comments.

News

• Nikos has prepared some new paperwork on deliverables, schedule, etc., for the ATLAS UK oversight comittee.
• Fred: UK agreed to pay its full share of the cost to complete. (ATLAS calculation was about 1.2M CHF higher than what the UK had agreed to contribute.) The invoice has been sent from CERN last week.
• Welcome to the newcomers to T2UK - joining the Manchester group.

Progress reports

• Andrzej on ROB-ins: 5 out of 20 something cards are broken and have been sent back from CERN to Royal Holloway. Four of these have been diagnosed with the same seemingly-simple problem: The card does not respond to communication thru the PCI bus, but the FPGA, PPC, etc. work as probed from the serial-connector backdoor. This indicates a problem with only the pci communication chip. Two of the damaged cards were sent back to CamCraft to be fixed (others kepts for further studies at RHUL). They were x-rayed in order to identify any possible mechanical problems, but none were observed. Later they were sent back to RHUL, after being "fixed", but they came back worse, with nothing working at all. Finally, CamCraft was asked to remake two new cards, after communication with German colleagues, in particular Andreas.

A little history on the cards: The cards were initially tested at the production site. Later about 10% were tested at RHUL. Once they arrived at CERN, each got tested for about 15 minutes, but this problem seems to appear on the long-term. So for UK cards have not been used extensively. In the coming weeks more of the UK cards will be used and then more concrete steps can be decided on. Cards had been temperature-cycled.

• Gordon on ROS software: TDAQ 01 07 has recently been released. Included in this release are quite a lot of internal changes, but most people will not observe the differences from outside. Now running on SL4 for tests, with binaries prepared with gcc345 instead of gcc323. It looks as if it runs slightly slower in SL4 than it does in SL3. Apart from that there is a problem is related to iostream: if a thread gets cancelled during a cout or cerr, the next access to iostream causes SEGV. As an immediate solution, all such outputs are to be removed from the opt version. An alternative solution that is being considered is to have a macro which seals thread cancellation when an output is to be made. In addition, for some trials in the pit with the ROS software running, machines froze and had to be powercycled. The ROS code used on those trials are believed to be free of memory leaks (since there is no dynamic memory allocation and since after hours of running the memory footprint stays constant and since the problem occurred in the beginning of the runs as well as later). It was possible to ping the frozen machines, but ssh did not respond. This issue is currently under investigation. Additional information from Fred: ROS group decided to buy a fraction of the new hardware with higher specifications, which might give a little help in the software speed.

• Fred on HLT nodes: TDAQ group has decided to buy dual-socket quad-core machines. 130 boxes are now on their way from DELL. (The unpacking/installing of the boxes will need some planning, possibly using cranes.) Each machine has 8GB of memory, extendable (in units of ~2GB) to 16GB. Given that an AthenaMT job consumes about 800MB and an offline job about 1.2-1.3GB, memory might or might not be tight (depending on how much library-sharing can be of help). For the 900GeV runs, it is possible to switch off some of the cores. Tests with multiple AthenaMTs show that 1.6GHz quad-core machines outperform 3GHz dual-core ones by a factor of about 20-30%. LHCb also did some tests, which favor high clock speed, but this finding might be related to how/if they use multithreading. Simon did some tests comparing the performance of the offline code on the quad-core machines with that on the AMD four-socket dual-core ones, and got results in favor of the quad-cores. Finally, some Woodcrest (dual-core) 3GHz machines have also been ordered, mainly for monitoring etc.

• Fred on networks: While 2007 additions are being planned, 2008 planning has also started. Because of logistics, it is better to think now about installing issues, when the HLT racks are being installed.

selection software reports

• John on tracking: Dmitry has been migrating IDScan to new steering and configurables. New work is essentially needed on the converters to go into release 13. L2 should use the same tools. Dmitry's timing measurements with release 12.0.5 are as good as the older measurements or sometimes better.

• Ricardo on truth: L2 truth is planned to go into release 13. While this will be mainly based on offline truth matching, some work will still be needed: a single truth particle is associated to a single track in the offline code, and more associations will be needed for L2.

• Mark on Regionselector: Started looking at RoI shaping improvements.

• Paul on monitoring: Monitoring tool is not in any release yet, but will be put in soon. Feedback from endusers are mainly about better choices for histogram binning and range selection. Idscan cosmic version went into 12.0.5-HLT. Online running of this will wait until mid-late summer. (Paul got the L2 ID tracking monitoring from James Keates and made the tool version of his algorithm. He helped to Jamie Boyd and J. Ferland to integrate it for the cosmic and jet slices. Also working on the conversion of xml files for online slices to newer format for use with confdb.)

• Simon on release schedules: Release 13 (and onwards) will be organized differently from previous releases. The different projects will go out on different dates, starting from 7th of march. For example, Regionselector on the 7th, and the Trigger on the 21st. Same sort of schedule is planned for release 14, but with longer spread. Trigger in release 14 is meant for sometime around july.

Items for release 13: Migrations that John mentioned (see above), BS converters, configuration, steering, and some tuning of the algorithms. There are a few changes to the event data. One issue recently raised is about the size of data per event. There are big overheads because the offline code makes copies of all trigger-related pointers. Serialization is being worked on. (ESD has been storing all the EF tracks, although this was never asked for.)

• Jiri Masik on trigger running: Event display for the control room is to be integrated in the next technical run.

• Simon Head: The L1 part of the Triggertools is in good shape, and now the HLT part is moving in. Database so far with mysql.
• Monica on documentation: She created/maintains the "Trigger User Pages" on ATLAS TWiki. https://twiki.cern.ch/twiki/bin/view/Atlas/TriggerUserPages This page is meant as a portal for people who would like to do Trigger-Aware analysis. Any useful information that can be put on the page will be appreciated.

physics

• Monica on egamma: Offline electron identification efficiency is about 70% with respect to truth. In order to tune trigger and offline performance together, feedback and iterations are now starting. Release 13 will have some of the updated hypothesis algos.
• John on B-physics: Julie has been looking at updated results on B-physics in the new release. There are some worries in the testing of the B-physics menu. There are new mass-constraint vertex fits implemented by Dmitry.
• Simon Head: Compared electron trigger efficiencies for tops and Zs. (See slides for details.)

future

• What is happening in the near future? Release 13, and the same sort of validation work. Improving the monitoring, above mentioned migration, new converters, the constant battle against memory leaks (particularly for the EF). Menus are becoming more realistic, with misalignment, configurations etc.
• Fred, Paul & Jiri on online tests: Online integration people would eventually like to have some algo-side people in the next few months, but probably not in the very near future, when they are focusing on more basic issues.
• Manchester is continuing its efforts in getting ready for the trigger running on the GRID.
• Ricardo (in july) and Erkcan (in march) are moving to CERN on LTA in order to work for commissioning.

finances

• Fred on finances: Travel budget of 30k, only used 11k so far. For short trips are allocated 15k - half is used so far. Finances for travelling in the next year (ie after april) are still ok, but the year after that will be tight.

• Bids: To be located in building 32, two machines have been bid for. In november we got a Woodcrest, and another dual-socket four-core 2.33GHz machine has been ordered. Bids are still open, but we should be quick before the end of the financial year. We are sitting comfortably, unless it turns out we have to rebuild all the ROB-ins.

next meetings

• Phone meeting: March 14, 11:00am.
• Face-to-face: April 24 at RAL, two days later ATLAS-UK collaboration board.

-- NikosKonstantinidis - 07 Mar 2007