Software Meeting Minutes
August 17,2004


D. Lawrence


Decisions from last week:

1. The XML schemas are the definitive source of in memory format info
   (as opposed to the hddm XML files)
2. Congratulations, I now pronounce you ROOT and GlueX.


Current Agenda:

1. XML Schemas and HDDM: What are they? What do they give us?   R. Jones

2. Dave's soap box of the month: "2ms! How much can we possibly do in just
   2ms!"


1.
Richard gave an abreviated discussion on what XML schemas are and what
benefits they have for using them to describe our data. It became clear
that XML schemas have been developed for describing data which is
precisely what we wish to use it for. In order to get the functionality
we want from a simpler XML format, we would need to re-implement
many features which already exist in available parsers. There were no
compelling arguments to implement something other than schemas.

2.
Dave should the results of a few memory speed tests he did to try and
measure the overhead incurred with large numbers of memory allocations
and deallocations per event. 
(See http://www.jlab.or/~davidl/HallD/MemoryTest.php)
The result was that we might expect to spend as much as 6% of our time
per event in memory management. This is based on 2ms per event which
will be required by the L3 farm from the design report. The 2ms must
include all I/O and reconstruction/analysis.
Dave suggested this motivates us to use an approach with HDDM in-memory
formatting which recycles memory rather than create/discard. It was
pointed out that 6% will not likely be the driving term when profiling
the fast reconstruction code. Dave argued his proposed plan has no clear
cost, but does have a clear benefit. It was agreed that we should keep
speed concerns in the back of our minds when developing reconstruction
code.