Cooked data file output: (Will, Stepan et. al.)
Will and Stepan presented an analysis of the BOS banks
produced in cooked data files. A consensus emerged that we will
write try to write cooked files which will contain the "uncooked"
banks EC, CC, SC, CALL, HEAD, TRGS, DCX(see below) and Hit and Time
based
tracking information in a compressed form(see below) plus some
seb output banks. The strategy is that Hit Based tracking takes
the majority of CPU time and is fairly stable, the goal of the
first pass analysis will be to do hit based tracking with the
assumption that the time based tracking will be redone at a later date
- In addition only events with one or more track will be written.
The above implies the tracking related banks
must be structured such that the results of hit based tracking can
be used to re-do time based.
Two sets of drift chamber banks
represented the majority of the data - they were the HBLA and
the DC1 banks (this issue was also discussed the previous week
- no minutes). The HBLA bank contains the hits the drift chamber
planes and is used by ced and drift chamber monitoring programs -
it contains a large amount of data (2304 bytes per track) much of
which is redundant or can be derived from other banks. Our strategy
will be to try to restructure the tracking code such that writing
this bank will not be necessary to re-do time based tracking. In
addition a constructor will be written to recreate this bank from
HBTR, HBTB and other banks if necessary for use in ced and other
programs - Jim Mueller will try to address these problems.
The DC1 one bank is about twice as large as the DC0 bank
this occurs for two reasons: First, the bank consists of an int and
a float which is twice as large as the two 16 bit integers stored
per hit in DC0 - second, very little compression is occurring due to the
very large tdc window cuts. One option is to not write out the DC1 bank
and instead write DC0. To do this we must then modify the tracking code
such
that it does not require the DC1 bank to re-do time based analysis.
Alternately, we can write out a new bank which consists of two 16 bit
ints
which is a mirror of the DC0 bank but with tdc window cuts applied -
this may save us some disk space - a more careful examination of
tdc window cuts may also save a large amount of CPU time - Joe Manak
will try to address these problems.
pdu - Dan
Dan has been using pdu and compared the list of dead, and hot
wires pdu finds with his own knowledge of dead and hot wires in the
CLAS. He found pdu to be 100% efficient in finding hot wires and
96% > efficient in finding dead wires - certifying that
pdu is a useful drift chamber monitoring tool.
D.C. Geometry - Reinhard
The survey numbers have been placed in the database. There
were a few problems last week with computer bugs that were straightened
out by Wednesday - the default settings in recsis give a width of the
elastic
peak between 16-20MeV for run 7867. There is some evidence that the
new time to distance functions (not the default) can give a width of
14 MeV for the same run.
Next Week -
Recsis/a1 timing - Kungseon, Joe
Database Management - Mac
tracking bank status
cvs update
- Joe