ScienceTools: (Jim) See the Science Tools blog in Confluence for the latest news.
FSSC: Not much news but lots of ambient noise.
u09: (Richard) u09 is filling up again. (Heather) will take a look.
Building here, running there funny things are everywhere (Jim) There are users with questions concerning using the RHEL4 32bit builds in batch: there has been trouble on the RHEL5 64 bit batch machines when using readline. Should we be telling them to restrict their jobs to RHEL4, or move to the SCons builds? (Heather) asked that Jim forward one of the messages, and maybe we can either provide a fix, or ask the Computing Division to install the older version of readline. Our intent is that we should be able to run the RHEL4 builds on the RHEL5 machines. (Tom) runs RSP in batch such that he checks the version of the machine and chooses the matching version. When the job is on a RHEL4 machine, he chooses the RHEL4 build, and when the machine is RHEL5-64, he chooses the RHEL5 32 bit builds. (Richard) was surprised by this, stating that in his analyses, he believes he is using the RHEL4 builds on RHEL5 just fine. He'll check his setup again. [Many thanks to Heather for covering the play-by-play during the early innings. ed.]
Documentation: (Chuck) sends the following:
It's business as usual in WB world. I made a new tag last week, so we're now at workbook-07-00-09 and, just to remind everybody, you can always check on the major changes by clicking on the 'LAT Workbook Change Log' link located just below the image on the WB's "splash" page. One thing that may become a concern is links to outdated pages in the workbook that may exist on other sites, e.g., a Data Portal page or a Confluence page. The most recent was to a Data Challenge 2 page on the WIRED Event Viewer. Unfortunately, there was nothing on the obsolete page itself to indicate that it was outdated, and people ended up wasting a significant amount of time. Of course, the only links in the WB are from the DC2 section, but these pages also show up when people use a search engine, including the one in the WB. So..., we may want to consider how to handle this type of situation going forward. One possibility is to simply flag each page that is no longer current; or, we could move them all to a site with a 'blind' link.
Other significant projects coming up include the ripple effect when we switch to SCons builds only.
Additional notes [thank you, Heather!]: Newsletter will be coming out next week. And Chuck disappears to Paris for two weeks starting next Thursday, providing British Air and the volcano agree.
GR: (Richard, since Anders' new laptop doesn't like EVO, or vice versa) Much of the C&A meeting yesterday was devoted to concerns about the CTBCore variable. It's causing significant differences when applied to MC and real data at high energies. This was already the case for Pass 6 but is even worse for Pass 7 since the cut value is higher. The discrepancy was discovered by looking at earth limb data which comes from the LEO era; it seems to point to mismatch between CAL and TKR. It therefore is possible that alignment has something to do with this. (Tom) has been asked to reprocess this data. He estimates it will take a couple days to a week to set up for this run, then about a week—if there are no significant snags—to execute.
(Tracy) See also Eric's further analysis, which points to gaps between towers as a likely significant factor.
Pass 7 news: (Anders) None to speak of.
Pass 8: (Tracy) Focus is chiefly on getting ready for the workshop at the end of this month: formulating goals and path to reach them. Otherwise, concentration is on getting Overlay in. Leon would have liked it in the merit n-tuple, but there is a cart-before-the-horse problem with that, so he is resigned to doing without. Work on incorporating errors in clusters continues.
GR versions: (Heather) Tracy found and fixed the major memory leak in RootIo (an unintended consequence of updates for WIRED) and Joanne made a new GR tag incorporating it last week. This fix should also be applied on the Pass 7 branch. (Richard) Even with the fix jobs are taking 2-3 Gbytes of memory, but we're surviving. (Warren) 32 bit jobs get 4 Gbytes on a 64 bit machine (on Linux).
Core Week (Heather) It looks like the original proposal, week of July 19, is the best after all, though not perfect. (Tom) Could it start a little earlier, towards the end of the preceding week? (Heather) That might work; she'll look into it.
Gaudi upgrade (Heather) Almost there! She's run into a glitch with test_GlastSvc: can't find the converter for McEvent, though it is able to find the one for Event all right.
flux (Heather) Riccardo made an update to flux 2 weeks ago but, in checking it out, discovered it doesn't fix the problem it was supposed to fix. He suspects another file in the package may need a similar patch.
L1 and ROOT (Heather) Newer tags on the L1 branch have the new ROOT (5.26). (Warren) will try it out. (Anders) expects to use the current L1 production version for the reprocessing, however.
SCons topics (Richard) is concerned that we demonstrate that Windows ST developers will survive the move to SCons builds only. (Heather) will poke Toby straight away. She also mentioned that we need to coordinate with the L1processing folks to make sure they can use the ft2Util SCons builds and adequately set up using CMT-built GR and SCons-built ST. Warren agreed to try out updating his scripts to set up GR and ST that way. He will also take another peek at ft2Util.
(Tom S.) has tracked down the intermittent problem with automatic creation of CMT tags following manual creation of an SCons tag. It's basically a race condition between the CVS server and the cron job which is supposed to discover new SCons tags and make the corresponding CMT tag. There was a small oversight in his first attempt at a fix (basic idea is to detect the failure and retry—a few times if necessary—to avoid being too early with the CMT tag attempt) but he believes he's taken care of it in the latest version. We have to wait for the race condition to show up again to be sure.
He also looked into differences in behavior of unit tests for 32-bit and 64-bit machines. He and Masa tracked it down to a difference in the representation of double in the two architectures. Fixes have been made so that test programs now succeed or fail identically independent of word size.
(Jim) would like to see email notifications from new RM builds. (Heather, Joanne) Navid had implemented this, then turned it off while SCons RM was being debugged. There had been a request for a more informative subject line on his to-do list which should be attended to before we turn notification back on.
(Jim) The SCons RM web interface sometimes reports good status for test programs which have in fact failed.
(Joanne) has no SCons news to speak of. Updating workbook documentation is her top SCons priority (until the next fire breaks out somewhere) but she has mostly been tied up with EXO recently.
(Heather) SCons 1.3.0 has been installed at SLAC for Linux builds; we could switch over to it any time. SCons Central has at this point also come out with a checkpoint release in the 1.3.0 series and an alpha 2.0.0.
previous | minutes index | next |