DØ Level 3 Algorithms Meeting: 20th February 2002

Talks, verbal reports

Introduction, p11.02.00 Aims and Timescales: Terry Wyatt.
The aim of the build managers is to get p11.02.00 built early next week. For L3 we expect to have implemented by then:
- a large number of bug fixes to the muon track reconstruction and filter,
- transverse shower shape cuts for the electron tool,
- much improved CFT unpacking.
There is some concern that with the huge number of bits of reco code that will be migrating from p10.15.00 into p11, it may take some time for p11.02 to become stable and fully functional. This would not be good for L3! Harry Melanson is aware that we need p11.02 urgently and, if necessary, he and Alan will do a partial build of the minimum bits needed for L3.
Gustaaf has been grappling all week with a 2.5 kByte per event memory leak in the p11.01.00 code currently running on the L3 farm. The behaviour is quite wierd. No leak apparent on IRIX. Very small leak apparent offline on Linux, which goes away when the electron tool is swiched off.
Two new activities:
- Marumi Kado will work on implementing the hot cell killer NADA into L3 and adapting it to the needs of online filtering.
- Andrei Zatserklyaniy is a new member of the L3 algorithms group. He is interested initially in monitoring the performance of the muon tool/filter. Welcome!
Muons:
- p11.01.00 performance/certification: Martijn Mulders
  The muon certification samples have all been processed without crashes, using a trigger file that includes also the calorimeter-based filters. The p11.01.00 tool finds muons that have been identified as "tight" quality by the offline code with an efficiency of 50-60%. A few items still need to be done to complete the certification process for p11.01.00:
  - Timing information will be provided by Christophe Clement.
  - Test samples of relevance to the calorimeter-based filters need to be run to check that the muon filter doesn't affect the performance of those filters. (As a general point: we will work towards having a common set of certification samples defined, with one set of jobs being run to produce rootuples that all tool authors can use. This saves effort and ensures that the certification is carried out for the complete trigger list.)
  - Persistency of the muon physics_results will be checked.
- p11.02.00 improvements: Martin Wegner
  Martin presented a long list of bugs that have been found and fixed. The bottom line is that for single muons |eta|<2 the efficiency goes up from around 50% to around 75%. Several of Martin's fixes should improve the performance also of the offline muon reconstruction. Martin will coordinate getting all of his modifications into cvs with the offline muon guys. This needs to be completed by Friday, so that we have the weekend to test the code as fully as possible, before the Monday deadline for p11.02.00.
- Memory leak:
  One of the remaining problems on the high priority list for L3Muon is the presence of a memory leak (at the rate of approx. 1.5 kB/event) in the muon track reconstruction code. This affects both offline and L3, but is a more serious problem for L3. This code was written by Frederic Deliot, who is unfortunately not available to work on the problem. In his absence, Boris Tuchming has been working at Saclay to try to locate the leak. Indeed, he has found and fixed a leak, but unfortunately it was not the leak. (After implementing the fix, the code still leaks at about the same rate as before.) Christophe, Boris and Paul have tried using purify, to find the leak, but without any success. Given that the muon guys have been struggling with this problem for some time, it would seem that we need help from someone who is expert in using the tools for tracking down this sort of problem. At Harry Melanson's suggestion we will contact Amber and Wyatt to ask for a suitable expert to help out with tracking down the problem.
- Unpacker:
  No progress towards getting the dynamic unpacker operational was reported at the meeting. Paul Balm, the author of the "old" unpacker, says that if a module is removed from the muon readout without a corresponding change to the cfg.dat file, then the only consequence is that the data for the associated muon readout crate will be ignored. Neither crashes nor garbage data are produced. A hole obviously appears in the acceptance for the L3Muon tool, but this is not a serious enough problem to hold up the muon filter going online.
A huge amount of progress has been made by Martijn and Martin, with the help of everyone who has been participating in the "L3Muon task force".
CFT Unpacking Progress: Robert Illingworth
The CFT Unpacker now uses individual channel thresholds. Daniel will study the effect on global tracking.
Electrons p11.02.00 improvements: Ulla Blumenschein and Volker Buescher reported that they are on track to get the code necessary for transverse shower shape cuts into p11.02. The current loose electron cuts will be retained for high p_t and also the possibility of a track match can be put in by a change to the trigger list.

Next week we will have an update on p11.01.00 data checks for taus from Yann Coadou, who will provide an "idiots" level guide to decoding the trigger information and comparing on- and off-line trigger bits.

Longer term things

Shadow nodes: receive a copy of the data that goes to an online L3 farm node and can be used to test new releases in a near to perfect simulation of "real life". We're realising more and more how useful this will be.
We need to worry seriously about how calibration and alignment constants will be input to L3 and versions properly tracked.

Scribe: Terry Wyatt.