Core Minutes 11/15/2011ADASS 2011, Paris (Richard) There were tutorials (as well as talks) on GPU's and R. Cloud computing was another hot topic. See also Richard's extensive notes, including several links to projects of interest.
ScienceTools: (Jim) The Binned/Unbinned task force that Elizabeth is chairing met last week, and Steve Fegan gave his presentation describing the tests he had done on the diffuse response integral calculations. He found that by simply switching the order of integration for the phi and theta dimensions (doing the theta-integrals first), he was able to get results that matched the slower, map-based integrals. I enabled those changes in a local ST build and ran some tests on the binned vs unbinned TS distributions. For the hard-sources paper, these distributions looked as if the unbinned TS values were overestimated. The results from my simulation studies are that the TS values I find using the new integration scheme agree better between the binned and unbinned cases, so this is promising. I am continuing with tests that Jean, Zhaoyu, and Johann had done back around the Saclay Collaboration meeting where they first characterized a bias in the counts attributed to the isotropic vs Galactic diffuse components.
FSSC: (Eric W.) is engrossed in the tip overhaul. Still some weeks of work to do.
Reprocessing: (Tom G.) Scripts and bookkeeping used for reprocessing have been accreting barnacles so he embarked on an overhaul which likely will be done by the end of the week. Pass 7 scripts have been migrated and are ready to go if needed.
(Richard) Stephen Zimmer will try using Tom's scripts at Lyon. We'll want this ability for Pass8 if not for Pass7. After any glitches encountered are addressed, we can move on to issues of data storage and transfer.
Disks" (Richard) Because of the flooding in Thailand, it's prudent to address disk needs sooner rather than later. We're purchasing 400 TBytes. Part of this will go to replace 100 Tbytes of old (5 or 6 years) nfs servers which have been getting flaky.
GR/Pass8 (Tracy) Everything is fine. We're aiming for a new GR tag at the end of this week. A few things still need to be committed. The new tag will have a new classification scheme for assigning energy to tracks and the new recon flow.
Discussions on truncation continue. Although there are more truncation events with conversion in the thin layers, the effect there is relatively insignificant. The same can't be said for thick layers, especially since more events there come in at a large angle. This issue will be discussed at this Thursday's Pass8 meeting and probably also at the C & A meeting next Monday.
Demise of redhat 4 (Heather) All externals are now available for redhat5 32-bit and We have (necessarily SCons) builds of GR Pass7 and Pass8 for redhat 5. We need a systest comparision of SCons versus CMT builds on redhat 4. Julie has offered to help with that. Then we should probably do some MC runs.
She's in the midst of merging Gaudi upgrades in the Pass7 branch. Pass8 (main branch) will be next.
ST and redhat 6 RM (Heather) All externals are ready to go and RM is producing LATEST builds. There was one more hitch: most test programs were failing with a complaint about the environment variable CALDB, even programs having no need of it. (Joanne) tracked it down to code in facilities which defines a bunch of path variables. It checks to see if a particular directory exists by doing a file open. For unknown reasons this test always fails on redhat6. But she found an alternate method, using the stat( ) function, which works on all Linux platforms. (Jim) Why not just go ahead and define the variable without any sort of check? (Joanne) If there is a supersede directory as well as a base directory this strategy would give the wrong answer for packages existing only in the base directory. (Tom S.) can add automated HEAD and tagged builds at any time.
SCons Windows RM (Tom S.) Since he moved the configuration from the registry to a file made the fix to ensure that the environment variable APPDATA was defined, all has been well, or nearly. There was a brief period when builds were failing after CVS checkout, due to an introduced bug, but that is now fixed. HEAD and tagge the problem. He noticed that builds would succeed while he was logged on via remote desktop, as if some token or key was available then but not necessarily otherwise. The missing bit appears to be an enivronment variable, APPDATA. He added it in on glastlnx20, which launches RM jobs. Since then (Friday) jobs have been running successfully.
obf (Joanne) made new-gaudi updates to all packages needed to build and run test_OnboardFilter (local only, in order to test the Windows build of obf B3-0-0). The first attempt was a miserable failure, but that may have been because I neglected to update job options files.
(Joanne) Tony W. and JJ have implemented or will implement various changes for the official release of B3-0-0 which should make our job of porting significantly easier.
Announcements
SCons meeting Wednesday at 10:30
|
|
minutes index
|
next
|