Core Minutes 6/5/2012
Core Minutes 6/5/2012ScienceTools: (Jim) reports via email
IGRF I updated the model in astro to IGRF-11. Leon's been following the progress, and I have a link to the summary at the ST Development page.
I removed the EarthPhenom model from the flux package. Its ROOT-dependence violates an agreement we had with the FSSC to limit ROOT-dependence in ST. Keith Bechtol is adding the new Earth Limb model a separate package under celestialSources.
FSSC: (Tom S.) has been working on a version of the Fermi Data app for IPhone and Android. He released it to FSSC for internal testing.
(Alex) There has been some success overcoming ROOT dependency issues in the current version of ST; it will be fixed in the next release. There was a question whether a dependence of gtbin on healpix would cause problems. We don't foresee any.
Pass7 reprocessing: (Tom G.) We're nearly there. ROOT production through March 31st is done. FITS file production is nearly done, just a few jobs to re-run.
New boxes (Tom G.) Lance Nakata has given us a proposed configuration to look at for compute servers: Dell R410 machines with 2 6-core ships, 48 Gbytes of memory. We'd like to get by with just 10 of these. Now that Wilko is back, we can get started on figuring out what to do about replacing the sulky's.
Disk (Richard) We have about 160 Tbytes left which has to last us to our next purchase. That should be enough for about 5 months.
Pass7 and Pass8: (Leon) The exercise of copying/converting new-truncation-style data to old blew up for both Pass7 and Pass8 because a tag containing a necessary patch didn't make it into the releases used. Will be fixed in the next round.
He's currently looking into a smallish memory leak (25 Kbyte/event, but for these runs, which process lots of events it mounts up) with Valgrind; no luck so far.
Pass8: (Leon, Tracy) Still dealing with remnants of new Gaudi fall-out. The latest problem had to do with auditor variables, some of which disappeared. Tracy managed to make them reappear, but they weren't filled. He now believes he can take care of that issue. Another recent problem had to do with Overlay and converter protocol, since fixed. Tracy notes the Gaudi conversion has held up Pass8 development by 2 or 3 weeks.
(Leon) noticed a second copy of IGRF code [see discussion in ScienceTools section above concerning the primary copy in astro] in CRflux. Help is solicited to understand why it's there and preferably get rid of it.
(Tracy) Johan is processing standard Pass8 datasets, used to examine recon. Since there were a lot of changes in need of validation, this is a large and necessary step forward, to be followed by work on background rejection.
SeeVogh: (Richard, Heather, Joanne) We had another go at it last week, still not too encouraging. Windows and Mac users were more or less ok though sound quality on Mac tends to deteriorate to the point where the application has to be restarted. Linux users had more serious problems. The suggested solution usually involves installing a new graphics driver, possibly also modifying configuration parameters or installing alternate versions of various software packages. I (Joanne) am not willing to go to such lengths and possibly disrupt the state of my rhel5 laptop, which is essential for just about everything else I do, just for SeeVogh. The Linux problems have to do with video and we have no need for high-performance video. (Richard) has made the point that it needs to work on Linux with no intervention, but thinks he detects a bit of denial.
Documentation: (Heather) tried out an etag on a particular page to get new workbook content to show up in browsers; it worked as advertised. We need to decide how widely this (etag) should be deployed and whether such deployment needs to be automated.
(Joanne) updated all three SCons pages. The system Tony J set up — commit to CVS to see pages show up in the test version of the workbook, run a script to copy to prod — is easy to use.
Additional instructions just for prospective SCons Windows GR developers are in Confluence. GoGui documentation could use some updating, especially for Windows users who wish to start Visual Studio from within GoGui, but I might be the only one who cares about that.
Core talk:
GR validation (Heather) CMT Systest results for the new tag 20-03-01 are not identical tag 19-04-01-gr17 (Pass8 branch) since the random number seeds applied are not identical. Tracy took a quick look and thought the results were probably acceptable.
CMT RM woes on Linux (Heather, Michael) For over three weeks CMT RM bulds of GlastRelease had been skipping certain packages. It turned they were precisely those packages which had changed since a particular build (HEAD1-1396) had been made. (Michael) discovered there were still remnants of the check-out for that build in the temporary work directory that CMT RM uses and this was the source of the problem. Heather was able to make good builds by hand during this period because the work directory in that case was different. Once Michael cleaned out the junk, new builds were handled properly. He has considered modifying the RM code to use a unique working directory for each build, but the risk of disturbing something somewhere else seems greater than the pay-off. This problem is clearly extremely rare, perhaps the result of a race condition. Michael will document.
CMT RM on Windows (Leon) Packages are failing to build, seemingly because they're trying to build in the wrong order. They need the CalUtil library and it isn't there yet. [update: it seems it's AcdUtilCommon.lib which is the problem]
GR vc90 (Joanne) The creation of rootmaps was wreaking havoc with SCons RM builds for vc90. For CRflux the genmap command would behave bizarrely, apparently executing code in astro. The execution would fail because certain set-up hadn't been done (because genmap is not supposed to be running user code!) and would tie up the build indefinitely. The command fails on Linux as well, but the build process moves on. Since this problem was preventing the creation of zip files for the Installers and since the rootmaps are of no apparent use, I took them out of the build. Recent LATEST builds and the new HEAD build (1396) are complete and available to the installer.
Next up: run-time problems. test_OnboardFilter fails in a very bad way. It took a while to find evidence that I had ever gotten it to work on vc90 with obf B3-0-0, but it's here in these very minutes. I later found further proof in a log of results from that time (mid-February). So most likely there is nothing terribly wrong with the Windows build of obf B3-0-0, but other than that I have no clue.
SCons RM (Tom S.) It's running all right; nothing to report.
rhel6 and GR (Heather) Johann has been working through the externals. He was able to build geant4 but ran into a problem with Gaudi and its use of ROOT which was later determined to be an issue between RHEL5-64 and RHEL6. Johann has since gone to working on the RHEL6 SLAC machines and has built Gaudi. He will not attempt to build GR.
Announcements
previous
|
minutes index
|
next
 |
J. Bogart, Last Modified: