GlueX Level-3 Trigger Meeting, Jun 24, 2016
From GlueXWiki
Contents
Meeting Info.
Meeting Time And Location
11:00 EST (JLab time)
CC A110
Connecting
You can connect using BlueJeans Video conferencing (ID: 198 332 417). (Click "Expand" to the right for details -->):
- Make sure you have created a BlueJeans account via your JLab CUE account using this link:
- http://jlab.bluejeans.com (You should only need to do this once)
- Meeting ID: 198 332 417
- (you may need to type this in, depending how you connect)
- If connecting via Web Browser: click this link (no passcode is needed):
- If connecting via iOS or Android App:
- Use your JLab e-mail address to log in and then enter the meeting ID given above to join the meeting
- If connecting via Phone: Dial one of the following numbers and then enter the meeting ID above and hit "#" or "##"
- Dial: 1 888 240 2560 (US or Canada only)
- or, List of International Numbers
- If connecting via Polycom unit:
- Dial 199.48.152.152 or bjn.vc
- Enter meeting ID above
Background Information
(Click expand for some links to background information -->)
Informational Links:
Justin's collaboration meeting talks:
Agenda
- Announcements
- Previous Meeting (Jun 10, 2016)
- EVIO Processing
- Benchmarking status
- Simulation Status
- AOT
Minutes
Attendees: David L., Simon T., Adesh S., Cristiano F., Curtis M., Justin S., Eugene C., Sean D., Beni Z.
Announcements
- Current meeting time overlaps with JLab seminar (when they occur). It would be preferable to move it out of conflict with this time.
- Time was chose to be same as DIRC meeting time so those involved with both meetings could have one at the same time every week.
- Some time constraints on DIRC participants from Germany and Indiana. Justin will suggest 9:00am on Thursdays. Regardless, they will determine a time and communicate it to us and we will likely shift to then as well.
Benchmarking Status
- Not much signifcant progress since last time
- Started looking at filtering out non-physics events when benchmarking. Preliminary results seemed to indicate no significant change in the profiling times.
- Still large discrepancy between processing of simulated and real data. Cristiano will continue to pursue it
EVIO Processing
- Steady state event processing rates in thread configuration space was mapped.
- Motivated by "Parsing Only" case where unexplained drop was seen from 12kHz to 8kHz
- Map shows a clear optimal region (~14 parsing threads and 18 processing threads) where close to 13kHz is maintained
- Similar exercise done for Wire-based tracking and Time-based tracking. Neither of these showed an optimal region
- Highest rates need full complement (32) of JANA processing threads
- Wire-based tracking required at least 3 EVIO parsing threads to reach maximum rate
- Will need to repeat exercise after MVA is developed to determine optimal thread configuration.
- Maximum single Ivy Bridge node rate (no parsing, no linking) was ~13kHz
- Maximum single Ivy Bridge node rate when Wire-based tracking enabled was ~600Hz
- Expect final algorithm to be at least as fast as WBT
- (this would require <50 Ivy Bridge nodes)
SImulation
- Issue with first round of L3 simulated data set due to FCAL calibration constant configration.
- Adesh has fixed the issue and is generating data set
- Eugene noted that we need to start looking critically at how well the simulation matches the real data. This will be crucial for any trigger simulation.
- Sean said that Mark I. is in charge of producing a large data set with Eγ>3GeV and the individual detector groups will bear responsibility for comparing it to real data.
- It was noted that Alex S. is working on more detailed L1 trigger emulation using data from the RCDB.
Action Items
- Identify new meeting time for DIRC meeting that doesn't conflict with JLab seminar (Justin)
- Post slides shown at June 10th meeting on benchmarking (Cristiano)
- Compare charged tracks from real data and simulation (Adesh)
- Test playback of real data using DAQ system in hall and specialized ROL (David, Adesh)
- Find and report on some early work on classifying real data events as reconstructable or not. (David)