Offline Analysis Software for the RHIC-Phenix Experiment

Paper: 351
Session: A (talk)
Speaker: Morrison, David, University of Tennessee, Knoxville
Keywords: analysis, development environments, world-wide collaboration

Offline Analysis Software for the RHIC-Phenix Experiment

D. Morrison
S. Sorensen, V. Perevoztchikov
Y. Zhao, L. Qun, H. Geng, Q. Liu

University of Tennessee
Knoxville, TN 37996-1200

for the Phenix Collaboration


Phenix is one of four experiments currently under construction at the
RHIC accelerator complex at Brookhaven National Laboratory. The
experiment will study very high energy collisions of heavy-ions with
the aim of demonstrating the existence of a quark-gluon plasma and
investigating its properties. Working toward this goal, many of the
experiment's total of roughly 400 collaborators will participate
actively in developing offline software for physics analysis. The
experiment will collect 200-300 terabytes of raw data each year, and
keeping apace of the data collection will require approximately 100
Gflops of continuous processing power.

The Phenix computing model specifies that the bulk of the data will be
stored in a hierarchical storage system at the central RHIC Computing
Facility (RCF). The sheer volume of physics data provides serious
obstacles to a traditional nano-DST per physicist analysis model, so a
strong emphasis has been put on a storage architecture that will
permit database-like queries of the physics quantities. The RCF will
also be the site of most of the initial event reconstruction, later
physics analysis, and data mining. Plans are also underway for the
creation of a major regional computing center in Japan which will be
responsible for all simulation tasks and performing CPU intensive
event generation calculations.

In this initial phase of development, effort has been concentrated on
implementing a distributed analysis system in order to provide the
aggregate processing power required to analyze the large anticipated
amounts of experimental data. Tight financial and manpower
constraints have made commodity hardware and commercial software
intriguing and necessary parts of the overall computing model. To
accomodate the widely dispersed nature of the collaboration we have
implemented a code development environment based on CVS and AFS. The
computing model also addresses a strong need for web-based and
client/server tools to provide individual physicists with access to
centralized sources of information, such as documentation and