Minutes for January 15, 2003
Vivek joined us by phone and wanted to leave when he was done,
so he requested to go first.
- Status of L1 CFT/CPS Simulation - CFT, CPSAxial and CPSStereo:
Vivek Jain
- PS of talk
- there is still a geometry discrepancy between the actual detector
and the MC that needs to be resolved
- d0trigsim Status/News: Dugan O'Neil
- PDF of talk
- Vivek's talk really didn't address CTT unpacking, but this
functionality has been added for real data
- 1,515 event crash on Wh
- ttbar MC sample used is inclusive
- crash points at a vector of stuff getting written into the
L3chunk, which is getting corrupted
- Josh Kalk sees a similar crash with 500 ttbar events
which he says points to L3 muon
- Jon Hays says the library used for this code snippet is picked
at random, so the L3 muon pointer may not be correct. One has to
dig into the D0om stuff
- if you turn off L3 then things run fine
- 3,410 event crash on ttbar sample with L3 turned off
- points to ReadEvent
- at the crash point the destructors just kick in,
so the tuple looks ok
- only occurs with a "debug" build, not a "maxopt" build
- will wait until have a working executible under gcc, then try
using ValGrind to hunt down these crashes
- authors are encouraged to get p13 fixed in addition to fixing the
test release / gcc problems
- memory usage while running has been looked at,
and it's not increasing
- there's a L1mu-L2mu communication bug
- needs to be sent to Jim Kowalkowski
- version of COOR that is running online has hand-edits.
Can we do what they do and keep up with them?
- it could be a while until TrigSim actually goes onto the farms,
but it's ready to go
- L1Muon Simulation: Rob McCroskey
- PDF of talk
- Josh Kalk has seen problems running L1Cal in passthrough mode
which look similar to the L1muo problems that Rob is seeing
- Rob: Try running it in two parts. The unpacker can mess up
if it gets run twice.
- the unpacker has been looked at to fix this "twice called" bug,
but there is not an easy fix
- Tsim_l1cal Update: Josh Kalk
- PDF of talk
- the thumbnail documentation is still crude
- "L1cal" is stuff that could have been triggered on online
- online calculates the MEt and triggers on it,
but does not save the info
- need lookup tables put in p13.02.00 to get same results
- there are plans to change this online
- this actually happens to all online global sums
- Terry Wyatt comments that he thinks it would be best
to save the online global sums
p13.07
- thumbnail code has improved
- a run number term has been added to determine what eta coverage
should be used
Joe Kozminski will be taking over tsim_l1cal,
starting essentially now
- Tsim_l1frm (framework) Update: Josh Kalk
- PDF of talk
- in the tuple output, parameters labelled with "STnn" have
a value of 0 if the event didn't pass trigger "nn"
and 1 if it did pass
- the biggest change is new code for L1L2 unpacker, which is aimed at
completion fpr p14
- L2 Overview: Josh Dyer
- PDF or PS of talk
- trigsim certification
- there are already trigsimmed samples available on d0mino
- need feedback on what plots to post to certify each tool
- Josh Kalk wants to know what the Ht cut is on - EM, jets?
- his point is that L2 doesn't differentiate between EM objects
and jets, which leads to some double counting
- suggest use of "only em" or "only jets"
- jets are now 5x5 instead of 3x3
- MJT filter added, which adds up the jets and gives opposite vector
- the L2 online certification procedure is currently being developed
- right now it look a lot like the trigsim process
- eventually it will be based more on the L2 global monitor setup
- aim is to make it more automated
- use only MC?
- keep reusing the same data set?
- Status of the STT Simulation: Todd Adams
- focus is on test release, not production releases
- have not tackled data errors yet
- t02.43.00 works, but t02.44.00 and beyond don't (bug still unknown)
- if you want to turn off L2stt, you need to turn it off as an input
to L2 global too
- right now the hardware and the simulation don't match, and its
the simulation that is correct
- width of the residuals is down to about 11 microns
- NB: these residual calculations contain the hit used,
so they are biased
- the reconstructed tracks are a little narrower than the generated
tracks. Need to check the logic of that result
- currently the STT simulation is the largest memory and CPU user in
trigsim
- note that this is just true for the first event when a bunch of
tables get loaded. Subsequent events are much faster
- TrigSim Status Report - L2MU: Christos Leonidopoulos
- PS of talk
- new central lookup tables, their first major upgrade since May, 2002
- cuts are slightly stricter than reco - considering loosening them
- the "correct" geometry for the forward lookup tables is in place
as of version p13.03.00
- "UBS" - a flag to save online inputs (= UnBiased Sample)
- L3 Overview: Jon
- PDF of talk
- certification for L3FMuon is emminant,
need to put last bits of code in
- will put p13.06.01 online to fix mark-and-pass bit setting
for special runs
- Discussion
- Josh Kalk: believe the solution of the event-number crashes talked
about by Dugan should promote a p13.08 build if necessary
Most people left at this point. Dugan, Jon, Terry Wyatt, and Angela stayed
to discuss the certification process.
- Certification
- Adam Y. has written L2 certification document.
We should have a look at it
- need to talk to Reinhardt about his plans/strategy
for L2 certification
- Marco has asked about thumbnail certification,
but no plans have been made yet
- need to be sure the examines run the actual analyze packages to test them
instead of running a separate "test" version or being
separate entities
- need to write up a procedure for the certification process
- how should we start running on data? Start now?
- get a data set. Use run 170247?
- run with -l3only
- use USEL1=yes
- use triggerlist 10.0
- use trigsim p13.06.00
Scribe: A. Bellavance (all typos and inaccuracies are mine)
Questions? Comments? Problems? : bellavan@fnal.gov