Dear Colleagues,

  
	Here are the minutes of the Dec. 15, 1998 Meeting of MONARC 
Architecture Working Group:



1) Sverre Jarp -- The processor outlook -- a report from PASTA time

	We heard a report from Sverre Jarp of CERN and the PASTA technology 
tracking group on processor technology evolution. The conclusions are 
summarized in

    http://nicewww.cern.ch/~les/pasta/run2/minutes/minutes_27nov98.html

and the associated full presentation may be found at 

   http://nicewww.cern.ch/~sverre/PASTA_98/index.htm

(These are also accessible through the Architecture web page)

	The talk starts with a comparison of predictions made in 1996 with the
realities of 1999. PASTA anticipated a slower increase in clock speed than 
occurred and a more successful exploitation  of Instruction Level Parallelism 
(ILP) than actually occurred. The net result, however, was  that the 
projections were reasonably accurate. This demonstrates that projections of 
this kind can be  trusted at some level but also illustrates the problems in 
relying on them too completely since the previous successful projection could
have been fortuitous. There followed an analysis of trends between now 
and LHC startup circa 2005. The projection for the cost of processing is 
about $1/CU for 'low end' computers and $2/CU for 'servers'. There are some 
clouds on the horizon. 0.1 micron technology could be delayed. Or anticipated 
gains through parallelism might not be realized. Or the world could simply 
lose interest in  faster CPUs and the money for R&D and for building the 
required fabrication facilities might simply disappear. Based on this, Sverre 
recommended that we use a factor of two 'safety margin' in applying these  
cost/performance projections.

	The discussion was mostly upbeat about the prospects that ample
CPU power would be available for the analysis of LHC experiments. We want
to thank Sverre and his colleagues in the PASTA project for their work and
look forward to future reports on other key technologies.

2) Status Report on Survey of existing computing architectures
   (subtask 1)

   Material collected for the report appears on

        http://www.fnal.gov/projects/monarc/task2/task2_meetings.html 

   
 
        Tim has collected information, not in all cases complete yet,  
            from ALEPH, DELPHI, L3, OPAL, NA45, NA48, and COMPASS. 
            He is still in the process of accumulating information.
            Tables may be found on the web. 

        Ian McArthur had problems transmitting audio but provided
             a report on the ZEUS experiment at DESY on the web.

        Vivian O'dell summarized the status of her survey of KTeV at Fermilab
             and her data is also on the web.

        Mike Diesburg summarized work on the survey of D0 and CDF at 
             Fermilab. He is relying heavily on the report "Joint CDF/CD/D0
             Data Management Needs Assessment" which contains detailed
             information about the CDF and D0 analysis experience from
	     the last collider run. This document is available at
             http://www.fnal.gov/wwwdocs/projects/runii/runiiweb/



        
3) Discussion of  the form of the report, the schedule, and the 
           format of key tables.
          
	There was a long and interesting discussion both with respect to the
status reports we heard and to the final report. 

	It was proposed to add three more items to the table that were more 
specifically geared towards analysis issues. These related to the generation, 
storage, and distribution of Monte Carlo event; to the generation of standard 
NTUPLES; and to the final analysis computing. 

	While this proposal received general approval,  it opened up the
whole question of the relation of this effort with the survey being
undertaken by the Analysis Working group. The chairmen agreed to discuss
this. It was also emphasized that that the report ought to develop
information from its tables and text that would inform and guide the
simulation effort. Some coordination with the Simulation Working
group is also required.

	Another major issue that emerged from the four status reports
related to the degree of distributed computing. One got the impression
that very little computing was done away from the Experiment host laboratory,
whether it was CERN, Fermilab, or DESY. Some people said that this was
misleading -- that CERN had good experience with major offsite centers such
as IN2P3. Fermilab also had successes in this area but not so much in the
collider experiments. D0, however, had good success in distributing the
responsibility for production Monte Carlo. The discussion of the difficulties
of actually distributing the analysis load identified several issues
including the problem of maintaining coherence between the activities, code
bases, calibration constants, and datasets at the various sites; the problem
of people at remote sites being 'out of the loop' -- that is not being able to
participate in the intense, ongoing discussions and interactions which
always go on at the main analysis center at the host lab; and the issue of 
support and maintenance capabilities at the remote sites. Since the role of
regional centers and the efficacy of distributed computing are main concerns
of this working group, it is strongly desired to include these issues in
the survey of past efforts. In particular, there is a need to collect 
information about the amount and kinds of distributed tasks that were
successfully carried out in the past and to identify the key parameters
whose values either made distributed analysis possible or prevented it
for being effective. Successes in the area of distributed computing
should be sought out explicitly by the people collecting data since,
as we've seen from the early efforts, they tend to get lost. It was clear
the leaders of the other working groups wanted to have strong input to the
direction and final form of the report and everyone agreed to that.


	After this, discussion turned to the details of producing the report. 
It was agreed to compose it in HTML to facilitate rapid distribution and 
response. After the text is finalized, the report will put in some standard 
form (such as WORD or TeX) whose choice will be left to the report 
editor.

	The plan is to pull the tables and other information developed
for the report together by the next meeting or soon thereafter and to have a 
draft for inspection by January 15. That will leave a week for comments
and revisions, which is not a very long time. We will do our best.

	The next meeting will be on January 7, 1999 at 16:00 MET.


							Sincerely,

							Joel Butler