GlueX Offline Meeting, June 8, 2016
GlueX Offline Software Meeting
Wednesday, June 8, 2016
1:30 pm EDT
JLab: CEBAF Center F326/327
- 1 Agenda
- 2 Communication Information
- 3 Minutes
- Review of minutes from May 25 (all)
- Spring 2016 Run Processing Status (Paul)
- Tagged releases of sim-recon and hdds: recon-2016_02-ver01
- simX.X (Sean)
- DTreeInterface (Paul)
- Review of recent pull requests (all)
- Review of recent discussion on the Gluex Software Help List.
- Action Item Review
- The BlueJeans meeting number is 968 592 007 .
- Join the Meeting via BlueJeans
Talks can be deposited in the directory
/group/halld/www/halldweb/html/talks/2016 on the JLab CUE. This directory is accessible from the web at https://halldweb.jlab.org/talks/2016/ .
You can view a recording of this meeting on the BlueJeans site.
- FSU: Alyssa Henderson
- IU: Matt Shepherd
- JLab: Alexander Austregesilo, Mark Ito (chair), David Lawrence, Paul Mattione, Nathan Sparks, Justin Stevens
- NU: Sean Dobbs
- Nightly Build Re-Config. Mark sent out an email describing the changes due to the change to GCC 4.8 or better. The line-up of nodes where the build is done has been updated on the wiki.
- Small Files on Cache Disk. Mark went over his recent email announcing a program of deleting small files from the cache disk.
- Sean has already started clean-up of files produced by production runs so far. Accounting files have been removed; ROOT files have been combined resulting in one file per run.
- The reconstruction launch is producing skims of synch events that are very small and are being written to cache. For now they will not be written to tape; we have some time to decide whether to move them or let them be written to tape from their current location.
- Dropping selected tags from EVIO data. Sean has checked in a change that allows this. He estimates that the reduction of the size of the π0 skims is an order of magnitude. This method has not been deployed yet. He is waiting for feedback from the calorimeter calibrators.
Review of minutes from May 25
We went over the minutes. Mark mentioned that he sent in the request to Sandy Philpott for the 200 TB of Lustre space we think we will need.
Spring 2016 Run Processing Status
Paul submitted the jobs for the reconstruction launch this morning.
- The first 50 jobs failed due to SWIF errors. Subsequent jobs started successfully. There are 200 jobs running now.
- There are 6500 jobs altogether. Each job requests 14 cores.
- There is no longer a mode where we can request an entire node in a way that is independent of the number of cores on a node. We should look at whether restoring that capability should be restored.
- Paul has tagged releases of sim-recon and hdds used for this launch: recon-2016_02-ver01
- Mark mentioned he has tagged sim-recon and hdds as well using the numeric scheme. Justin suggested those tags be brought into alignment with the launch tags.
Sean has put together sim1.1 a set of conditions for sim1.1. The plan is to run similar statistics to sim1. This data set will be used as input for Paul's efficiency and resolution plug-ins and the results will be compared with real data to obtain corrections for mcsmear. He has identified the collimator size (3.4 mm or 5 mm) as a condition that should be simulated separately.
He proposes using real run numbers and distributing the simulated data in proportion to the distribution in real data.
Matt also suggested we think about simulating signals with appropriate run dependent conditions.
Sean proposes another set of simulation, with higher statistics, incorporating what we learn from sim1.1. As part of this[?] he would like to see us institute some form of channel-by-channel, hit level efficiencies. He has already started designing CCDB tables to institute such a scheme.
Paul cautioned us that faithful simulation of detector efficiencies might be more complex than independent channel-by-channel random number generation. We agreed that that was generally the case, but this scheme provides a good starting point.
Access to the RCDB during reconstruction and analysis
Mark is working on a scheme to add RCDB to the standard software suite so that we can take advantage of Dmitry Romanov's C++ API to the RCDB.
Mark reminded us that the RCDB does not have a history mechanism. It is not possible to set a time-stamp to recover a previous incarnation of what the RCDB thought the running conditions were. Previous versions are only available via backups. We might want to enhance our backup system to provide for easier user access to past versions of the database.
Paul took us through his email announcing a new interface for filling ROOT trees where best-practices locking is done transparently for the user. Users do not deal with ROOT objects directly; all access is done through the interface. One caution is that the user has to make sure to destroy the interface object at the end of the job so the interface knows when to close the ROOT file.
There is an incompatibility with the Apple-provided version of the Clang compiler. Nathan has finessed this in the HDPM system.