Core Minutes 21 April 2015

Core Minutes 4/21/2015

Present: Joe Asercion, Joanne Bogart, Jim Chiang, Richard Dubois, Warren Focke, Tom Glanzman, Heather Kelly, Michael Kuss, David Landriu, Leon Rochester,

ScienceTools: (Jim) The FSSC wants a change in the gtselect behavior. This was motivated in part because we deliver FT1 and extended FT1 files with event class selection DSS keywords already present, so users making subselections to a cleaner class need to be able to overwrite the existing keywords.

Since the event classes are no longer necessarily nested, additional information needs to be available (via CALDB) that specifies which event class down selections are allowed. Matthew provided that information, and I will incorporate the desired functionality into gtselect.

This work will involve branching the version of the dataSubselector package from the version included in ST-10-00-03, since the later development in ST-10-01-00 includes changes to the external libraries, the addition of wcslib and healpix, that the FSSC thinks would take too long to vet for the forthcoming release.

See Science Tools Development Notes for current activity.
FSSC: (Joe) No news other than the request Jim described.
Reprocessing: (Tom) Backfill through April 7 is complete. This may or may not be the penultimate backfill.
Hardware: (Tom) Jon Bergman is looking into options for setting up quotas and so forth on the new gpfs servers. Once this is determined, we'll be able to move over. Meanwhile, much of the copying has already been accomplished with background rsync.
C&A: (Leon) Verification and consistency checks continue. One remaining problem concerns PSF fractions. Front-back is ok, but quantiles are not yet right: more events end up in PSF0 (best predicted PSF) and fewer in PSF3 (worst) for real data as compared to MC. Leon has an idea about a possible culprit. The tracker walls are represented as a solid uniform material, which is not realistic; there are channels, etc. The amount of material real tracks go through is much more complicated than this simple model suggests. He says [or, rather patiently wrote up at my request]

My idea is to run the MC twice, once with +x% material in the walls, and again with -x% material, and then generate the predicted PSFs based on using, say, a 50/50 mix of events of each type. So the average material would be the same. This would simulate a model where half the tracks went through less material, and half through more.

If this generated an effect in the right direction, the % and mix fraction could be tuned to get a better answer.

People are thinking about this.. it may be possible to rule this idea out without doing all the hard work, based on what we know already.
Pass8 target date: Apparently not yet determined. (Warren) reports that M.E. appears to be satisfied with L1 readiness
CHEP report: (Richard) It was very LHC-centric. He concentrated on the Cloud sessions since those were most relevant for us. Mostly they concerned OpenStack; Docker might also be of interest. You can find links to all the talks at the Conference site. See Track 7 for Cloud talks.

Here at SLAC SCS had been using a product called Nebula, based on OpenStack, to provide VMs for testing and development, but Nebula just went belly-up. SCS will probably move to another OpenStack-based product; there are several. We expect to be depending on such a thing to keep GR (at least) going; most likely we will not advance beyond rhel6.
Core talk:
- GR news (Heather) There is a new GR, 20-10-02, incorporating newer versions of astro and evUtils already used by ST. See details on the GlastRelease Updates page.
- TKR monitoring and Pass8 (Heather) M.E. discovered an incompatibility in data structure expectations between packages involved in TKR monitoring (calibTkrUtil and related packages) and Pass8. Hiro has agreed to work on it. Michael will provide support as needed.

minutes index

J. Bogart, Last Modified: