Core Minutes 8/7/2012
Core Minutes 8/7/2012ScienceTools: (Jim) A variety of fixes are done or in progress:
The Workbook documentation in CVS has been updated for gtltcube and phi-dependence. Getting this update out onto the web is on hold until the issue with Hudson and the workbook is sorted out. (and the Hudson builds are now working again) See Science Tools Development Notes for details.
Hardware: (Tom G.) xroot storage: Five systems have been ordered (450 TB of space). The servers (Dell R610) have arrived, the disk arrays may not arrive until late September.
NFS storage: Four wains (Sun Thor-class machines) each with 32 TB of space are being repurposed from xroot to NFS service. The first (wain025) is nearly ready to take over the Fermi user disk areas; a cut-over is being planned for Thursday 16 August (concurrent with the power outage). The next, wain026 continues to be drained. Richard mentioned that the 500 GB request from the dark matter group is on hold until the user space is expanded.
glastlnx replacement: Ten new Dell R410 machines have just been ordered to (hopefully) replace all 25 of the existing glastlnx machines.
Still looking into whether Fermi contribuion to batch farm is consistent with the propsed PPA contribution prepared by Stuart Marshall. This is to replace 400 cores of our batch allocation.
The issue concerning filtered merits and the astroserver has been discussed at the data handling meeting. Tom has a Confluence page devoted to the topic.
Pass7 (Leon) Not much news. He's been working on two calibration projects: new calibrations for L & EO and automated calibration creation. Sasha will be producing more calibration files soon.
Leon spent a pleasant day working on SCons on Linux, which seems to work, though he doesn't always understand what's happening. Now starting to look at Windows. Visual Studio 2008 has been installed, and next up is GoGui.
Tracy commented on the simulation results for Riccardo Rando to make new IRFs using the software truncation which seem not to be producing the same answer. Leon responded that the software truncation is complicated in that you want it to do its thing and not affect other bits...then when you add Overlays there's the question of whether it has been transcribed correctly. Richard asked to remind the listening audience about the core issue, this is the idea that in software you were going to replicate the truncation, but have the simulation results look the way it did before? Leon went on to remind us of the general idea: In the new readout configuration (Trunc64), we allow the GTRC (plane) buffers to fill out to their max (64) hits, which means that for some events the upper layers will be truncated because of GTCC (cable) buffer overflows. So we're trading better fidelity for high-energy events at the bottom of the tracker for some loss of hits further up. I turns out that this gives a net improvement in effective area and PSF, especially in Pass8. But it means changing the Level1 data products on the fly before we have the appropriate flight IRF's, and also mixing pre- and post-Trunc64 data, which seems like a bad idea. Bill and Tracy came up with the idea that after doing trunc64, we artificially truncate the GTRC buffers in software back to 14, so we should recover the old Trunc14 events in most cases. This way, Level1 can go on as before, but we will have the better data 'in the can' for use with Pass8. The initial checks in MC indicate that we don't get exactly back to the original; how serious the discrepancy is remains to be evaluated. Johan commented that there seems to be a 5 to 10% difference in the Aeff and PSF in some parts of the phase space which needs more investigation.
Pass8: (Tracy) The new version of GlastRelease is not completely out of the woods yet. There was a problem with the MC generation having to do with the environment variables for the start and end times not being picked up correctly. Matthew Wood discovered this issue. 1/5th of the events were missing in the SAA, where there seems to be some interplay between the endtime and when you're in the SAA in FluxAlg. Hopefully this is just a JO issues. It is not fatal to the simulations, but producing less events than normal, everything else is looking pretty good.
OBF: (Tracy) Tony Waite visited Tracy to discuss the need for RHEL4 support of OBF versions between B1-1-3 and B3-0-0. The actual gamma filter has been stable since a couple of months since launch. We discussed who might be able to help answer whether we need RHEL4 versions of other OBFs. Richard suggested C&A, and Johan asked that an email be sent to their mailing list. (that email was sent after the meeting)
Core talk:
RHEL5 validation (Heather) The system tests for GlastRelease 17-35-24-lp29 and 17-35-24-gr35 have been re-run with the additional merit tuple quantities that Johan has requested. The last re-run was to pick up the modifcations to the binning and range of the plots. Johan plans to look at the new runs soon.
SCons RM (Tom S.) Running smoothly.
Virtual machines (Tom S) No addtional news since last week. Johan confirmed Luca can get the VM working on his laptop but his desktop gets stuck. Possibly due to a linux kernel issue.
vc90/SCons developer environment Tracy reiterated what was covered last week concering the installed headers on Windows. Joanne has started to implement the modifications for windows only, and will complete them once she returns from vacation.
previous
|
minutes index
|
next
 |
J. Bogart, Last Modified: