CMS Computing Model

Paper: 200
Session: A (talk)
Speaker: Pimia, Martti, CERN, Geneva
Keywords: ?


CMS Computing Model
===================

CMS Collaboration

to be presented by

M. Pimia
CERN / ECP


Abstract

The CMS Computing Model represents the architecture for a
system to deal with managing and organizing four separate but inter-
connected sets of resources: our computing and networking hardware;
our data; our software; and our people. This must be done in the
context of more than a billion physics events per year, more than a
thousand physicists located at over more than one hundred institutes,
with a detector and physics complexity unprecedented in High Energy
Physics. Moreover, the long lifetime of the experiment coupled with
the rapid pace of technological advancement adds to the complexity by
requiring our model to be flexible enought to respond to changes in
technology. Computing, including software and networking, has become
a full detector subsystem.

Summarizing the key features of our proposed solutions:

* Hardware: Our hardware solution is characterized by large collections
of CPU, disk and robotic mass storage at CERN and a small number of
regional centers. We do not give precise specifications for the
division of hardware now, but rather a recipe for optimizing the
distribution of hardware at a later date based on hardware
availability at different CMS collaborating institutions and
international network costs and available bandwidth. In any case,
the hardware arrangement will be based on the principle of moving
the computing tasks to where the data resides, and providing
satisfactory access to CMS physicists, no matter where they are
located.

* Data: Our data access solution is based on a new paradigm.
Instead of relying on serial access to increasing selective data
sets like DST, mini-DST, etc. we will access all data (raw,
reconstructed, and physics data) from a common object store through
the same mechanisms. Our data storage systems will automatically
optimize data storage based on access patterns to provide efficient
data retrieval for CMS physicists throughout the collaboration.

* Software: Our software solution is based on a hierarchy of types
of software, with a professionally engineered framework into which
physics modules may be inserted; on a greater reliance on software
engineering to cope with the increasing complexity of the software
systems; and on new modern programming languages and methods which
provide tools for solving our software problems, in particular
for now C++ and Object Oriented Programming.

* People: In order to make use of the talents of our highly dispersed
collection of physicists, we will make use of new methods for
managing software activities, for code management and distribution,
and for collaborating at a distance. These methods may require
greater discipline than software activities have required in the
past, but the payoff in increased productivity for the experiment
should be well worth it.


A final key element of our plan is to manage the transition
from current programming practices to the new languages, programming
methodologies, data access methods and collaboration tools that will
be in widespread use when the experiment runs but will need to be
creatively phased in during the period of detector construction as
detector design, simulation, and test beam activities continue to
require continuous computing support.