Core Minutes 8/14/2012Hardware: (Tom G.) We have a purchase req for the glastlnx box replacements; the order might have gone out already.
A new nfs server for user data is ready to go. Once it's done everything (e.g. mount points; quotas) should look the same to users. The migration will be completed in conjunction with the power outage on Thursday which will eliminate the possibility of stale mount problems.
Two more nfs servers are almost ready. One (wain26) will hold code, such as RM builds (but not the CVS repository). The other, not expected to come online until after the Collaboration Meeting, will be used by ISOC ops and flight opts. See this document for details on all new nfs servers.
PPA will contribute $600K to the batch farm. Some machines are earmarked for Kavli and hence require the pricy infiniband interface. It's still TBD whether all machines will have it.
450 Tbytes of xroot space is on order, to be delivered on about Sept. 25th.
Workbook: (Heather) Automated Hudson builds of the Workbook (which had been failing) were temporarily stopped and restarted with a new log file. This has fixed the problem.
It seems no one uses the Workbook print version. Since it's difficult to maintain (every change to a "regular" Workbook page needs to be propagated to the print version as well) we're thinking of abandoning it. This will involve small changes to all the pages which have links to the print version.
We believe the master copy of the picture gallery is on Chuck's old machine. Richard has suggested that Karen export it into afs space. Then Heather would be able to add pictures from the barbecue last week.
Pass7 (Leon) Only news is the backfill task Tom is running. (Tom G.) It covers data from April 1st to July 31st and is running smoothly. L&EO reprocessing is on the horizon, awaiting improved calorimeter calibrations for that period. This is of special interest to the 130 GeV line folks. For this reprocessing we reclaimed some of the batch allocation we had lent to LSST. When we're done they can have it back.
Pass8: (Tracy) Expecting a new tag soon. (Heather) was holding off to sort out GlastClassify unit test issue, which is probably not new. (Tracy) Traditionally GlastClassify has built a static library and stand-alone application linked to it, and has also built a component library for use within GR. The libraries are derived from much of the same source. It's a complicated arrangement, but probably not worth fixing; better to move to using TMine.
RHEL4 decommission (Heather) We need to schedule shutdown of RHEL4 CMT RM builds. It would have happened sooner but for Heather's and Michael's recent absences.
Julie says JJ contacted her about the converion of GRB tools to rhel5/rhel6. The task is well in hand and she is delighted.
RHEL5 validation (Heather) Johan took a look at the RHEL4 17-35-24-lp29 versus RHEL4 17-35-24-gr35 system tests and noted that the shapes of most of the histograms of interested looked similar, but some of the low-level plots had a surprising difference in the number of entries - due to different numbers of digis, or MC hits. Heather took a look at the logs of the runs, and reminded ourselves that while event based seeding was used - the events generated are not identical. This can be seen by viewing the types of triggers in each run, as well as the number of each type of source in the case of the BackGndMixDC2 run. No one has commented further. Since then, Heather has run a RHEL5 17-35-24-gr35 versus the RHEL4 version and Johan took a quick look. He noted that there were surprising differences in the plots of VerticalGamma300GeV. Heather is trying a re-run of that particular test to see what happens.
vc90/SCons developer environment (Joanne) is close to done making code changes to support not installing headers on Windows (GlastRelease only). Progress and description of procedure may be found in Confluence. The amount of work needed for this one-time only conversion is manageable; she's more concerned about maintenance. When a developer modifies or adds code impacting compile-time dependencies on headers, Windows builds may break until appropriate changes are made to some xxLib.py file. Ideally all developers, but especially Windows developers, need to understand how this works. Examples can be found in the Confluence page.
SCons RM (Heather) The Windows distributions SCons RM makes end up with the wrong file protections so that the Installers can't access them. Up till now the only solution has been to change protections by hand. Tom S. (currently on vacation) is looking for the underlying problem.
DB migration final(?) notes (Joanne) The last access issue — failure to SELECT from the Archive database — was apparently due to uppercase/lowercase confusion and is now resolved.
Pre-migration the production mood database was backed up to glastlnx01 via a cron job using mysql root userid for that machine. It started failing as soon as the mysql database on glastlnx01 went away. It's been re-incarnated to run on glastlnx20, mirror to mysql-node03, and use the glastops mysql user.
|
|
minutes index
|
next
|