10/05/98
Report from the GLT FDR committee

Rowan Hamilton, University of Iowa
Ray Larsen, SLAC
Krista Marks, LBNL
Helmut Marsiske, SLAC


Executive Summary:
==================

The committee is impressed with the progress that has been achieved
on the hardware, but finds that more thorough testing is still 
necessary.
The committee is very concerned about the GLT schedule with respect to 
the cosmic run and finds it not realistic as presented, primarily 
because of crucial software that is lacking. 
The committee strongly recommends that BABAR/trigger management makes 
available more resources to assist with software development. 


Hardware:
=========

The hardware board itself seems quite solid as far as it has been
possible to test. One board is in test and a second is loaded close to
being ready to test. Both are needed for the cosmic run.

Tests are needed for: basic functionality of lookup tables and
memories; margins of frequency, voltage and temperature; and using
simulated MC events in which multiple events of different characteristics
follow one another closely.
This should be done via standard memory and margining tests; test of the
board with simulated MC events for a number of different events, and enough 
events to get statistically meaningful reliability results.

The DAQ hardware needs to be exercised as soon as possible. This is the
one area that has not been exercised on the board. All other tests have
indicated that the physical board is flawless, but until DAQ testing is
done, it is unclear that another fabrication of the pc board won't be
necessary.

Much more thorough tests of the GLT's various memories are necessary. 
In particular, a bit-by-bit test of all of the look-up-tables should be
conducted before any GLT board is validated. Also, as pointed out above,
a read-write test of the LUTs is not sufficient, since it accesses a 
different hardware path than that used in executing the algorithms at 
run time. Thus, it vital to conduct much more extensive MC tests, using 
many more events from different MC samples, and, most importantly, using
 many different sets of configuration data.

There is a need for a complete and detailed test plan for the three GLT
boards to be delivered.  This was not presented at the FDR.  It should
include the following items:
   (1) The statistics that will be obtained before the boards are considered 
       fully commissioned.   
   (2) A plan to successfully run thousands of MC generate events from input
       memory to output memory.
   (3) A plan to pass both back to back events (~10 or as many as the input
       memory will accommodate) and overlapping events through the board.
   (4) A plan for robust testing of the LUTs (10^9? read/write cycles).


Software:
=========

While impressive progress has been made in the hardware development 
of the GLT, there has unfortunately not been equivalent progress in 
the software development. During the review, targets were described
for installing one GLT board in IR2 on Oct 10, and having two available 
for IR2 use by Oct 31. This is an ambitious schedule in terms of
hardware alone, and an unrealistic one when combined with the need for
software development. It must be emphasized that the GLT is useless for 
the Cosmic Ray Run without DAQ software to support it.

Whereas the Feature Extraction code for the GLT should be fairly trivial, 
the configuration code could entail some sophisticated design which has 
not yet even begun. The LUTs for the GLT are large, and down-loading must 
be implemented in an efficient manner, including support for editing 
existing LUTs in the GLT. Also, this code has to be written against the 
APIs provided by DataFlow, and it appears that the GLT Group does not have 
enough experience to do this in the brief period before the Cosmic Ray Run.

Thus, the committee thinks it is necessary that the GLT Group obtain 
assistance in developing their software, so that those who have experience 
with the hardware can continue more extensive hardware tests while the DAQ 
software development progresses in parallel.


Schedule:
=========

The plan to complete the GLT to "full functionality" and install two
boards in BABAR by the end of October is unrealistic, because of incomplete 
testing software and insufficient time to conduct all the needed tests.
The software resources needed to complete all the necessary work do not
seem to be under control of the group which reported. Therefore, tight
scheduling is not possible.
The overall plan and schedule should be revised to reflect reality,
and as a contingency, the group should figure out a workaround or 
partial solution to meet the cosmic test requirements.
If it is possible to structure the work to bring more resources to
bear, this should be done to bring testing to final completion as
expeditiously as possible.