Core Minutes 2/5/2013
Core Minutes 2/5/2013ScienceTools: (Jim) The meeting with Steve (Fegan) and Emilia last week was productive. Emilia's first project, starting last Novemeber, was to improve integration schemes. She presented her work so far and will finish it up soon. On a longer timescale, plans were made to extensively refactor the Likelihood classes for both binned and unbinned likelihood. Jim wrote a new base class which has been committed to CVS along a branch. Steve will be working on the unbinned classes. (Richard) Steve has committed to long-term ST support, a heart-warming development.
See Science Tools Development Notes for latest ST news.
Reprocessing (Tom G.) There is no reprocessing news at the moment, though Pass7 and Pass8 tasks are expected to start up before too long, possibly causing a disk crunch, especially if Pass8 output is significantly larger than before. (Richard) has suggested to Tracy that we could probably handle a 10-20% increase, but not much more.
Hardware and migration report (Tom G.) The new lnx boxes have been configured with 3 VMs per physical machine. Each of the VMs has 4 cores and 16 Gbytes memory at its disposal, more than the old glastlnx machines had.
Three xroot servers (sulkys) have been retired. Wilko, in the process of moving over some very old MC files, reached the inode limit on the new servers. We need another small partition for inodes on each of the new Dell servers.
Pass 8 (Tracy) Last week's GR tag, 20-08-06, successfully fixed some serious blocking issues. See JIRA LPATE-109 for details. The plan is to make a new GR tag this week which will write out tree-based tracking output (see JIRA LPATE-98 so we don't have to regenerate it. To start with, will output all of it. We'll see how much it adds to output and will determine what we can do without. Then there are plenty of other issues to attend to.
(Leon) Heather gave him an assignment to see if the latest version of tskim works with the new ROOT. This is something he needs for the tskim job he is working on to routinely create overlay events in a Pass8 context. Stephen Zimmer will help with pipeline aspects.
Michael Kuss will be taking on the 2-calibration issue.
Core talk:
GR and CVS (Heather) was able to create the newest GR tag, 20-08-06, without incident by patiently waiting for LATEST builds to complete first.
New flight release (Joanne) Made a new obf external (B3-1-3) for redhat5 32-bit. Because of the nature of the changes from the previous version, the new version didn't take much time to create; redhat6 64-bit should be even faster since the steps are the same once a couple parameters have been established. The new external has not been tested. She is hopeful no code changes will be needed and this will go very quickly. [Not so! several code changes will be needed after all — routine ones which shouldn't add too much time to the process, but will add a disproportionate amount of tedium.]
Which Mac OS? (Heather) Results of the Doodle poll lean towards Mountain, but there there are several people who claim to use Lion and a few who took the "Please don't take my Snow Leopard away" option seriously. This size of the response is bewildering, given the lack of feedback we've had regarding poor Mac support (that is, problems providing installable builds) to date. (Tom S.) FSSC skipped Lion. They still provide Snow Leopard builds, but most likely will drop it as soon as all issues with native Mountain Lion builds are resolved. (Richard) The upgrade from Lion to Mountain Lion is not difficult. (Heather) will confirm with Yemi that we can go to Mountain Lion (not the case some months ago). (Richard) Then ask him to help spec out two boxes. Presumably they'll be MacPro's. We'll need to make decisions about memory and disk and about where to put them.
Mac, RM, and Jenkins (Tom S.) As of last week Tony J. had set up Jenkins on bldmac01 to make Mac RM builds (that is, execute the RM programs for checkout, compile and test one after the other). All that was missing was automatic triggering. Tom has now set this up by adding a new Workflow to RM which triggers Jenkins rather than submitting jobs to lsf. The new Workflow is currently being used just for for ST LATEST builds. All signs are that it's working properly: he deleted a particular LATEST build and it was remade by the new Workflow this morning. We could consider using local rather than network disks for the build but it doesn't appear to be necessary; performance was fine with network disks. We should go to this scheme for Windows as well since we've had so many problems with Windows lsf, but first he has to learn how to set up Jenkins on Windows. We might also ultimately employ it for Linux, just to increase uniformity and thereby simplify maintenance.
PPA Cluster (Tom G.) The cluster (redhat 6, 64-bit machines) is up and running. There is a an interactive pool (bullet) consisting of the first two machines, and a batch queue (bulletq).
TBR
previous
|
minutes index
|
next
 |
J. Bogart, Last Modified: