Minutes from the Analysis Tools Meeting 2002 August 06. In attendance:
Adam (scribe), Heidi, Jianming, Jeremy, Jonathon, Michelle, and Peter

1) Adam generated several stream schemes and showed some slides based
on what he learned: the process of determining a stream scheme seems
to be handling the trade offs of keeping the number of physical
streams the W/Z group has to look at small and keeping the number of
streams everyone else has to look at small as well. One can't easily
do both. If the W/Z streams are given top priority, then everyone else
will potentially have to look at those streams. If the rare streams
are given top priority, then everyone (including the W/Z analyses)
will have to look at those streams. Granted those streams will be
small, but a lost file in a small stream amounts to losing many
luminosity blocks. That could have a big effect on other analyses that
need to look for a few events that got diverted to those small
streams.

2) Peter stated that an important feature of high pt lepton physics
(W/Z, top, higgs) is having the W and Z's as signposts. You can check
your top or higgs analysis by scaling in some way the number of W -> e
nu events. You can also self-normalize the W -> e nu events if they
are all in the same stream (that is not have to worry about
bookkeeping errors across streams). Even if bookkeeping errors are
supposed to be very small, self-normalization may be hindered if high
Pt lepton events are split among different streams. But regardless,
having high Pt leptons as the top priority streams would benefit W/Z,
higgs and the Top analyses greatly.

3) I showed some of the stream schemes I was playing with -- some
split the data up into many streams, some protect the W/Z analyses by
putting their data in one or a few streams, others are very unfriendly
to W/Z. The one we seemed to like the best was one that involved
adding primal streams to separate the high pt and low pt lepton
triggers, single and multi lepton triggers, and single and multi jet
triggers. So we can easily put the high Pt leptons into their own
streams. Heidi likes putting high pt E + X into its own stream (this
includes overlap with high Pt MU) and then high Pt MU + X into its own
stream. For the muon people, she suggests writing an inclusive E+MU+X
(only 0.8% of the global_CMT8.0 data) stream from reco. Low Pt leptons
would go to their own stream (split e and mu?). If the low Pt physics
analyses are truly orthogonal to the high pt physics (is that really
true?), they wouldn't have to look at the higher priority high pt
lepton streams at all.

4) Jianming and Jon would like to know how bad would life be if we did
ALL streaming (no physics streaming). How difficult would it be to
reprocess data. How does it affect remote institutes? We didn't get
back to this.

5) Heidi and Jon talked about keeping track of what triggers went to
what streams in the runs database. This would be useful to determine
if you really had to look at a stream that could have accepted your
events. For example, if the high PT leptons have the highest priority
and you are doing an analysis with a (contrived) 6 jet + missing Et +
tau trigger, it is possible that one of your events came in with a
high PT lepton and was diverted to that stream. But what if in reality
that never happened? With the runs database, you (or some tool) could
check this and know not to look in the high Pt leptons stream at
all. After some discussion on implementation ideas, Jeremy said he
could just do it.

6) With the realization that high PT lepton streams are important, the
"apolitical" stream scheme generator is being put aside. And in fact
it looks like some kind of priority scheme is indeed the
best. Michelle will begin working on spiffying up the interface to the
stream evaluation tool.