Core Minutes 6/14/2011
Core Minutes 6/14/2011Farewell, Chuck (Richard) Chuck has retired as of last Friday. We wish him well, and will do our best to keep his fine work intact and up to date.
Fermilab: (Richard) was there last week for an instititutional review. They are undergoing a thorough retooling, from the old collider-based program to neutrinos and such. These are interesting times.
ScienceTools: (Jim) 1) There has been progress on LLE tools. 2) There is a new EnergyBand class in Likelihood, making it possible to do analyses on different bands from a single FITS file. See Science Tools Development Notes for details.
FSSC: (Eric W.) has been dealing with the galactic diffuse model, new IRFs. (Jim) The diffuse model is large, which has been causing some grief in pipeline jobs. Has it caused problems as FSSC? (Eric) It's too big for CVS and it increases size of tarballs significantly but otherwise, no. (Richard) Tom G. saw process memory blowing up briefly to 2 or 3 Gbytes, (Jim) It's worse than that: a large percentage of jobs are crashing in seemingly random places. Jim ran one such in the debugger, both debug and optimize versions, without crashing. [See also Tom's report below.]
Pass7 Tom G., who couldn't make it to the meeting, kindly sent the following:
Reprocessing Last Tuesday it was decided to make two changes to the Pass7 reprocessing: 1) add diffuse response columns (Galactic and Isotropic (EG)) for 'source' and 'clean' event classes; 2) redefine the FT1 and LS1 file contents so that FSSC received only 'source class and above' in the FT1 and 'transient class and above' in the LS1. However, all photon event classes should be retained for the astroserver at SLAC. These two changes have meant some updates to bookkeeping and the creation of two new file species: EXTENDEDFT1 and EXTENDEDLS1 for use at SLAC. Work continues on implementing these changes and testing the results. Once this is complete, the entire science mission dataset will be reprocessed (from existing MERIT files). Slightly longer term, it is hoped to switch Level 1 to Pass7 and to release the data to the public by late July 2011.
gtdiffrsp I am concerned about the reliability of gtdiffrsp and the Pass7 diffuse model. Rerunning on the same small test run yields about a 20-25% failure rate (floating point exceptions). Different machines, different places in the code, etc. gtdiffrsp is taking ~3.1 GiB of memory during its setup phase, then drops down to ~1.5 GiB during event processing. This is a potential problem for batch machines which allocate a nominal 2 GiB per core. Experimentation shows that huge performance differences result if machines begin to thrash for swap space. This issue has not been resolved.
Pass8: (Tracy) GR v19r1p0 is imminent, but currently test_Gleam is failing on Linux while initializing G4Generator [as of end of meeting, no longer so. Heather tried it with Tracy's latest updates and it's all good. v19r1p0 is building.]
Johan had a request to investigate upward-looking gammas in the LAT which in turn led to a request to be able to create MC events with double overlays. The infrastructure is not set up for this. It would take about a solid week of work to implement but would probably be useful in other contexts as well. (Leon) Bill has suggested looking at (astro)physics involving coherent photons. We currently have no simple way to simulate multi-particle events.
Recon work is in its final stages. Cal is very stable; some lesser improvements are in the works for Tracker.
Dead and hot strips: (Leon) We're still running with pre-launch information. In fact very little has changed since then. One issue to be addressed is how to make data and MC more similar for partially-connected strips, of which there are ~1000 in tower 0. Maybe best to just kill them for data.
More GR (Heather) needs to do some retrofitting on the v17r35p1 branch (Pass7) and v15r47p12.
The gambit to replace G4HadronSim with code dependent on a Fortran package, to be incorporated via f2c, is on hold; G4 gurus have warned against using f2c. They would be delighted if we translated the code to C++, but resources for that are not forthcoming. (Tracy) Someday we might want the functionality provided by the package, but it can wait.
Core talk:
Disk clean-up (Heather) GLAST_EXT quotas which were too small have been extended. Kim has backed up rhel3 externals. As soon as we confirm the files can be retrieved from backups the originals can be deleted.
CMT news (Heather) Weird, random hangs have been occurring in Windows CMT builds; there seems to be no pattern. This symptom may be correlated with the addition of glast-win04 to the CMT RM pool.
ft2util_2 review (Heather) It went well. Giacomo did a generally fine job of preparing code, doc, and a review presentation. Links to all this and to Jim's detailed comments can be found in Confluence.
SCons RM (Tom S.) On Friday, after a lack of hue and cry, ST vc71 builds were turned off. Now that vc90 builds are undergoing more scrutiny, it's evident that test programs have been failing, in particular those dependendt on st_app. Heather will investigate on her XP machine.
GRBAnalysis (Joanne) GRBAnalysis-scons is nearly complete, lacking only the new subpackage LATBA, added to the old GRBAnalysis just last week. (Jim) LATBA can just be copied in; no need for SCons to install any of the scripts, etc., it contains. (Joanne) Nicola snd Vlasios have been working from the new container. Vlasios has also taken on the job of keeping GRBAnalysis and GRBAnalysis-scons in sync until such time as all users move to the new container [thank you, Vlasios!]. I made a HEAD tag yesterday. The output from tagCollector looked good; still need to confirm all the right stuff got tagged. Nicola has asked for a way to regularize the extra setup beyond that provided by the generic _setup file; work is in progress. See this Confluence page for updated plans and status.
Announcements
SCons meeting Tomorrow (Wednesday), 10:30 AM PDT
AOB
previous
|
minutes index
|
next
 |
J. Bogart, Last Modified: