Description of D0 Monte Carlo Production at CBPF
================================================

The system for remote Monte Carlo production is described in 2 CHEP97 papers. 
The core system based on a Client/Server interface:
   WWW Oriented Remote Job Submission Monitoring and Management over Internet
   -Proceedings of CHEP97 - Computing in High Energy Physics, 
    Berlin, Germany,7-11
    April 1997 - CBPF-NT-003/97, FERMILAB-CONF-97-140-E
 
and the farm for event processing:

   The LAFEX Risc Based Farm Solution for CPU Intensive Applications in
   HEP - Proceedings of CHEP97 - Computing in High Energy Physics, Berlin,
   Germany, 7-11 April 1997 


> We would be interested in how many events were generated, what machines you
> used, how much CPU time events took on average, how you made the output
> available, etc. 

We have generated in the last 2 years over a million events for several
analysis in D0. The machines, described in the Farm paper are Power PC
workstations from IBM, although this system could work virtually on any
kind of machine. CPU time/event depends on the event type. A typical 
QCD 2 jet event takes about 6 minutes while a top event over 10 minutes. 
The output is made available via a standard web interface running on a 
central server that keeps the job outputs for a finite period, after 
notifying the user that the job is done.

	We have to take into account the following points:
1. when we finished the farm it was exactly when finished the main  
   analysis in Dzero and the Monte Carlo needs and production decreases
   a lot. We had a lot of free time without job submission.

2. 	Our design had many targets: as easiest as possible to users
   submit jobs; maximum of independence of managers; delivery the 
   events at "home" users; maximum of flexibility for several hardware;
   and so on; use internet; and so on...

3. 	People had a big difficulty to use this facility, "it was too
   good" to have all these features. The time schedule did not coincided
   with begin strong needs of Monte Carlo Production.

	The machines of the farm production are: (now old machines)

Model: IBM RISC 7009-C20
Quantity: 1
Processor: POWERPC 604
Clock: 120Mhz
Memory: 256 M
Disk:  20 GB
OS: AIX 4.1
SPECint95: 3,85


Model: IBM RISC 7248-132 (43P)
Quantity: 30
Processor: POWERPC 604
Clock: 133Mhz
Memory: 192 M
Disk:  1 GB
OS: AIX 4.1
SPECint95: 4,72

Model: IBM RISC 7011-25T
Quantity: 1
Processor: POWERPC 601
Clock: 80Mhz
Memory: 128 M
Disk:  2.0 GB
OS: AIX 4.1
SPECint95:


Model: IBM RISC 7030-3CT
Quantity: 1
Processor: POWER2
Clock: 67Mhz
Memory: 256 M
Disk:  9.0 GB
OS: AIX 4.1
SPECint95: 3,41

Model: IBM RISC 7248-132 (43P)
Quantity: 1
Processor: POWERPC 604
Clock: 133 Mhz
Memory: 192 M
Disk:  2.0 GB
OS: AIX 4.1
SPECint95: 34,72

	What I would like to do now, is a farm of new powerful PCs
and get much more power, more efficiency, cheaper, and the software
is OK.  With the experience that we got with this project we saw that
it is possible to build a "Virtual Supercomputer" for HEP based in 
the principles of the national/internacional collaboration, using
the  farms and machines from different sites, adding the power 
installed in several places. If we succeed to convince people to do 
that we will have a real globalization of the Monte Carlo production.
This would be a nice project. A lot of software can be developed for
this and we have tools that we did not have in the recent past.

	The information above I got from Mariano and mainly from 
Gilvan who is at Fermilab in the present and if you need more 
information just give a call to him. (Gilvan Alves at Dzero
gilvan@fnal.gov)
	I will be at Fermilab between 9 and 16 of April. Let me 
know if this notes are OK for your ends.
	My best regards, 
				Alberto