The main topic of this meeting was to discuss the draft report 
on existing architectures. 

	We began with a discussion of some general comments on the report. 
It was observed that the report was intended to identify what you could
learn from this  generation of experiments and what you could not. Harvey
observed that it is clear from the report that the scale of the computing 
problem in these older experiments is so much smaller than for LHC experiments
that they cannot possibly DEFINE a framework for the LHC. However, the survey 
will help us make the best possible use of the experiences of the past in 
formulating what is essentially a new model of computing. We should state in 
the conclusion section that, at least based on this survey, a distributed 
computing model for the LHC will have to be developed as a qualitatively and 
quantitatively new kind of system in the course of the LHC project.

	Mauro observed that we can learn much by contrasting the way in
which Monte Carlo event generation and data analsyis were handled. First, 
remote sites made major contributions to the Monte Carlo work whereas they
did not for the most part contribute to the event reconstruction or to the 
data analysis. While the reasons for this were discussed briefly, Vivian
will amplify the discussion. Second, Mauro observed that even at central
sites, Monte Carlo and analysis tended to be done on different machines.
He wondered why this was and whether in a new distributed model it would be 
possible, desirable, and even natural to use all machines for both purposes.

	We then went through the document section by section.

Scope of the Survey: 

Perhaps it would be more polite to say that 'NA48 and NA45 are 
intermediate in  size of computing task although smaller in the size of
the collaboration'.

Harvey recalls that the first estimates for early LEP and early Run I were
low by one order of magnitude for the startup and by two orders  for
steady state running. This can be documented by comparing the numbers in
our tables to the 'Green Book' circa 1984. Tim Smith will try to locate a
copy of the Green Book. From memory, we believe that the early LEP estimate
was for 72 MIPS for analysis and reconstruction vs perhaps 20,000 finally
installed. The estimate for the 1988/89 FNAL collider run was 50 MIPS which was
off by a factor of at least 10 and is tiny compared to what was eventually
required. We should also recall, based on our previous discussion that
there are new drivers of computing requirements, such as the object database
overhead, which add to the need. Monte Carlo needs have also traditionally
been underestimated. The Monte Carlo needs are frequently of the same order
as the real data analysis.

Methodology of the Survey:

It was suggested that we add to the list of experiments and sumarizers
the 'contact person(s)' that they talked to. As long as the people agree, I 
think this is a good idea.

Quantitative Results of the Survey:

Concerning the survey numbers, Tim reminded us that the numbers in the
tables represent steady state numbers of the solutions that finally worked.
In some case, other approaches were tried that are not reflected in the
survey.

It was impossible to get people to provide numbers for 'event analysis time'.
These times varied greatly depending on the analysis and the sources were
reluctant to provide 'average' times. Tim suggested that maybe people
would be more wiling to give a range of times. The summarizers will try
to ask their sources on the experiment to answer the rephrased question.
Tim also suggested that we leave the 'sparse' columns in the text for now
since he still has hopes of filling them in.

On the question of non-central analysis resources, Laura remarked that
the kinds of machines available for each experiment among the Italian
research institutions can in fact be learned through the funding agencies.
It was felt that we should also leave the offsite CPU column in the 
report even though it was mostly blank.


It was decided to include the FOCUS experiment at FNAL in the survey. FOCUS
ran in 1996/97 and has used wide area networking along with tape export
to implement a distributed DST production. The major participating sites are
connected to VBNS. Joel will get the required information.

Other examples that people could think of  where systems have been used for 
remote computing are a system at Bologna which did Monte Carlo work on
wokstation clusters, the ACP system at Michigan State, D0 facilities in Brazil
(CBPF), Texas, and elsewhere. These are examples of offsite response to 
specific critical needs of an experiment. We need to try to get information
about these efforts.

Analysis and Conclusions:

People liked the idea of comparing the LHC requirements with those of this
generation in a table. However, it was noted that the test claimed that ALEPH
was in the table and it is not. It was decided that we should include 
representative or even maximal numbers for the LEP experiments rather than
single out one of them. It was observed that the BaBar and Run II nmubers are 
also small compared to the LHC numbers. That is the subject of another 
report.

We need to include words mentioned at this beginning of this note as
a conclusion. I repeat them here 

	It is clear from the report that the scale of the computing 
	problem in these older experiments is so much smaller than for LHC 
	experiments that they cannot possibly DEFINE a framework for the LHC. 
	However, the survey will help us make the best possible use of the 
	experiences of the past in formulating what is essentially a new 
	model of computing.

People accepted the recommendation for a method of tracking the next
round of experiments systematically. Joel will, however, come up with
a better title for the section.

	There will be a revision of this document issued early next week. 
People are asked to read the revised document and to send comments to 
Vivian.

Action List:
1) Expand the analysis of why Monte Carlo was more successfully distributed 
   than other parts of the task (Vivian)
2) Locate a copy of the Green Book and look up initial LEP estimates (Tim)
3) Get permission and list names of 'contact persons' on experiments --
   (survey team)
4) Get 'range of times' for analysis tasks (survey team)
5) Contact Laura to learn whether we canget a list of resources available
   for analysis at Italian institutions (Joel)
6) Add FOCUS (FNAL E831) to the survey (Joel)
7) Try to lean about D0 Monte Carlo production at CBPF and elsewhere (Mike)
   and Michigan State ACP farm (Joel)
8) Add paragraph above into conclusions (Vivian)
9) Generate revised draft (Vivian)