Attendance:
Jolien Creighton, Patrick Brady, Duncan Brown, Ed Daw, Greg Mendell,
Erik Katsavounidis, Siong Heng, Stuart Anderson, Alan Wiseman, Peter
Shawhan, Kent Blackburn, Masha Barnes, Warren Anderson, Joe Romano,
John Whelan.
Actions:
Alan Weinstein will e-mail group his vision of the summary
information needed for job summary/control.
Reports on actions:
Proposal to have the IFO and channel names in the domain structures
has been put into LALWrapperInterface. A new element to hold the
comment field from datacond is also in place. The new headers are
in CVS, but no functionality is available yet. Masha and Ed will
e-mail group when available.
All non-contentious issues about the modified frame spec have been
accepted by the SCCB. The local time and one other issue remain,
but will be worked out between the framers and VIRGO scientists.
Kent indicated that these changes will not propagate into LDAS until
the entire spec if ratified.
Agenda items:
Job summary/control:
--------------------
Alan Weinstein led of discussion about job summary/control
information. Things he would like easily accessible for each job
that is running in LDAS:
1. Job stats:
start time,
end time,
real-time duration,
LDAS installation in which this job was run,
user who submitted the job,
number of beowulf nodes requested/obtained,
elapsed CPU time in the DatacondAPI,
WrapperAPI,
any measure of disk usage,
IO load,
memory load, etc that might be available (to help optimize
the job performance),
name of input frame files,
name of DSO,
CVS tags of LDAS, LAL, LALwrapper.
2. Uniform input for classes of DSO: For example, in addition to
the data and its descriptors, a burst DSO might need:
mean and std of the input GW data stream (and skew, kurtosis?);
band-limited RMS noise in the GW channel, from [1, 1000] Hz,
from [200, 300] Hz (best response)
Is the IFO response function present and available?
Some measure of whether it makes sense
(eg, its magnitude at a few frequencies)
what, then, is h_rms from [100, 200] Hz ?
Is the expected noise power spectrum present and available?
Some measure of whether it makes sense
(eg, its magnitude at a few frequencies)
what, then, is S/N from [100, 200] Hz ?
3. DSO specific data:
values of significant filterparams, and any other internal
quantities that might change, such as threshold settings
number of sngl_burst triggers generated
anything else?
Following Alan's outline of issues he sees, there was some
discussion of what the right place for these things is. No
conclusion was reached, although it was generally agreed that
the information is available (or easily produced), but there is
currently no mechanism to collect it together. The issue was
summarized as need for a common loging mechanism.
Other business:
--------------
Overhead of running jobs in LDAS: Ed raised this issue for his
DSO. Basically, it spends only a small fraction of time
grinding data in the wrapperAPI. He was looking for a way to
stream jobs through the wrapperAPI. After considerable
discussion and clarification, the conclusion is that you cannot
do this with LDAS as designed.
Scaling LDAS to large Beowulf clusters: Stuart asked again about
this. No new information was available. Patrick and Duncan will
work to get something for the next meeting.
Finally, Patrick mentioned the issue of getting the CVS tags
into the database for discussion at the next meeting.
Kent asked that the agenda be circulated in sufficient time to
allow thought about the issues; Patrick apologized for the
recent tardiness
Next Meeting:
10:30 PST on Thursday, 7 February 2002
|