Date: Fri, 02 Apr 1999 17:24:13 -0800 (PST)
From: Marjorie Shapiro <shapiro@kfesg.lbl.gov>
Subject: A starting point for discussion...
To: bauerg@fnal.gov, chenyc@fnal.gov, cranshaw@fnal.gov, jbk@fnal.gov,
        kirsten@fnal.gov, ksmcf@fnal.gov, mark lancaster <lmark@kfesg.lbl.gov>,
        paus@fnal.gov, PCalafiura@LBL.gov, rharris@fnal.gov, wolbers@fnal.gov
Cc: patrick@fnal.gov, sexton@fnal.gov
Message-id: <Pine.SGI.3.96.990402171831.2998A-100000@kfesg.lbl.gov>
MIME-version: 1.0
Content-type: TEXT/PLAIN; charset=US-ASCII

I am forwarding a copy of an Email message Jim Kowalkowski sent out in Dec
summarizing some discussions he had with Kevin and Kirsten on Level 3
downloading of database constants.  The note contrasts the offline (read
user analysis for offline) access patterns to those in the farm and
discusses one possible solution for farm downloading where constants are
put in flat files that are shipped to the nodes.  Of course, no decisions
have been made about the access method, but I thought it would be useful
for people to read this message before next week's meeting and come to the
meeting with an opinion about whether or not this architecture is what you
want.


			Thanks
			Marjorie


Date: Wed, 16 Dec 1998 22:39:17 -0600
From: Jim Kowalkowski <jbk@fnal.gov>
To: Jack Cranshaw <CRANSHAW@acdft1.fnal.gov>,
    Kevin McFarland <ksmcf@fnal.gov>, KIRSTEN <KIRSTEN@FNALD.FNAL.GOV>,
    Marjorie Shapiro <MDSHAPIRO@csa6>, Dennis Box <dbox@fncdug.fnal.gov>,
    Mark Lancaster <lmark@kfesg.lbl.gov>,
    Liz Sexton-Kennedy <sexton@b0nd18.fnal.gov>,
    Paolo Calafiura <PCalafiura@LBL.gov>
Cc: Paul Marciniak <PAULM@fnal.gov>
Subject: Notes from Level4 DB meeting today...

Hi everyone,

I just want to briefly summarize what we discussed in the
meeting we had today concerning Level3 calibration database
access.  This is to make sure my understanding is correct (from
a high level) and for the benefit of the people who were not
present.

-----------------
Offline Summary:
The current offline architecture (planned) is basically a set of
database servers which AC++ processes attach to (as clients).
Each database server manages a connection to the database.
The server process loads calibrations from the database into
the server memory on demand from the client AC++ processes.
The server caches the calibration for a period of time (basically)
so the database access only occurs once for many accesses from
AC++ farm node processes.  The AC++ client processes
also cache the calibrations in their memory for repeated access
by module code running within the program.  A diagram here
would be nice - but this is supposed to be brief.

This is a demand driven system.  The cache is filled in the AC++
process as well as the server process as calibrations
are needed by code in the AC++ program.

If we peek into the running AC++ program, we see a calibration
managment system.  This calibration management system has children
which are really the I/O packages.  You would basically see one
I/O package per component/attribute pair in use.  Each I/O
package has a cache available for holding calibration objects in
use.  The calibration management system allows high-level functions
such as "flush the cache", and "set default calibration (calset)".
The user gets access to the data in the cache by using
attribute-managers.  There is one manager class per attribute/
calibration object.  The user uses a get/put interface of the
attr-manager
class to manipule data in the cache/database.  If we peek into
the running server process, we see the exact facility - the server app
is designed to deliver calibration objects to AC++ calibration
management systems.  It tranfers data to I/O packages in the
AC++ process.

------------------
Online discussion:
After a set of calibration data is made available in the database,
a set of validator processes is kicked off.  These processes
prepare calibrations required by level3.  When these processes
are done, they kick off a job on each of a set of Level3
gateway nodes.  The job on each of the level3 gateway nodes
is an AC++ program.  The purpose of this job is to cause
the calibration management system to be loaded with all the
"current" calibrations required by the level3 farm.  This will
required some sort of input file listing the required
component/attribute pairs.  After initialization, the AC++
process will have a complete set of "current" calibration objects
in memory, held within the calibration management system cache.

The AC++ process now issues a command to the calibration
database management system - "save to file xxx" (a new feature).
This causes all the calibration objects held in all the caches to
be written to a single file called xxx.  The intention here is that
a second command can be issued to another
calibration management system in another AC++ called
"restore from file xxx".  This command will presumably restore all
the objects as they were in the process that issued the "save".

The AC++ process now transfers the file xxx to all the farms
nodes that it knows about and puts the file in a well-defined place
with a well-defined name.

When the farm node AC++ process starts at the being of a run,
this file will be there for it to digest.  All the calibration data
it needs will be available - no database or server access is
required while it is running.  It would be desireable for
the file to be put into shared memory so only one copy per
node is required and multiple executables on the same node
can access the same copy.

This differs from the offline in that there is no real "database
server".
A set of programs create files, ship them to farm nodes, and exit.
The "process" is driven by calibration data being written into the
database.

The "save" feature of the calibration management system does not
currently exist.  Some thought must go into doing this in a timely
fashion.  Some calibration management system utility may need to
be created to allow forcing available I/O packages to be created
and load the "Current" calibration that they control access to.
A second solution may be to consider using SQL to prepare the
files from the calibrations in the database instead of an AC++ program.

Please fill in any missing parts of the picture (Kevin, Kisten, Jack).
I realize that this text is not very easy to digest, but it is getting
late.  Is there anyone I missed on the distribution list of this email.

Jim

--
Jim Kowalkowski   - jbk@fnal.gov

D0:  630/840-2215 - 5th floor of the D0 assembly building
CDF: 630/840-6457 - 169F in the CDF trailers

WWW:
  http://www-cdf.fnal.gov/internal/people/links/JamesKowalkowski/my.html

  http://www-d0.fnal.gov/~jbk/index.html