Attendance: Igor, Matt, Stuart, Albert, Peter, Kent, Scott,
Duncan, Bruce, Jolien, Patrick, Joe, Warren, JOhn, Ben, Philip,
Mariallessandra, Xavi,
Actions:
1. Bruce to start ball rolling on validation of detector
response function and barycentering code in LAL
2. Everybody is to start moving generic routines to a common
location in LAL. This is a long term task, but should get
started to clean up LAL. This needs some discussion. Jolien
will spend some time and circulate the organizational structure.
3. Get started with a basic set of instructions and guidelines
for LALApps development which can use condor. Try to answer
the questions that were raised about the why's and the how's.
Minutes:
1. Status LDAS:
they have been able to fix core dumping in the frame. Still
having some issues in mpiAPI, but still not ready to push a new
release. Cannot push anything before next week. Only if there
is pressure to do so. The sites appear to be doing pretty
well; unless that changes, will not push new version. New
version on ldas-sw next week.
Database deadlock. Configuration changes to reduce deadlock
problem by re-configuring DB2. Don't know if this is a real
problem. Alan's bug report on this. Kent will send Alan an
e-mail about this.
Development of 0.7.0 will be mid-late May.
2. LAL/LALWrapper
Julien still has a bug in TFclusters. Never converges. Will
change to bisection algorithm.
FFTW new release. Philip has looked more than Kent. There are
issues of changing the packing order. Kent cannot gaurantee
anything until a closer look is take at the new versions.
Jolien thinks the re-write of LAL would not be too difficult.
No transition to new FFTW until the S2a or S3 run release; i.e.
wait until about 6 months from April. October as target date to
transition to fftw3.
Seems like no real reason to upgrade to new version of fftw2.
Nor reports of numerical calculation.
Validation discussion at LSC meeting and how to proceed:
testing of antenna patterns. Greg did some independent testing
of Fx,+: no documentation. Did not test the timing issue.
Bruce will get the code that rejean used to test the barycenter
routines into lalapps. He will talk to some people about this.
Stuart asked about barycentering code: there are corrections
that may differ from the predictions. Do update right after
S2 is over.
Frame reading code: need to be able to do random access. The
new framelib may allow use of the TOC to do proper random
access. There seems to be a problem with getting the header
information. Jolien will try to resolve this.
Suggestion by Stuart to help people to figure out what needs to
be moved ....
Deploying search codes on grid:
Peter wanted to know what the scope is. Bruce has one
suggestion: concentration in LSUG on application. Lots o'
discussion about the technical aspects of running standalone
codes using condor and needs on various systems.
How about data movement. LDR will handle larger issues of data
movement. LDR uses cataloging approach which lists data
available at that sites. Query ...... what data. Sites update
each other with their contents. Local query. This
functionality will hopefully provide standard interface to the
codes through LAL, etc.
Stuart also asked why we were working toward the Grid at all.
Doesn't condor and LDAS solve our problem already? No
satisfying answer was presented except the generic one that the
grid will allow access to computational resources that are not
directly available.
There was an extensive discussion of the requirements for being
able to run codes at different sites. In particular, the
discussion of static versus dynamic linking came up.
Meeting adjourned at 13:40 CST.
|