PEP: the Phenix Event Processor

Paper: 257
Session: C (talk)
Speaker: Sorensen, Soren P., University of Tennessee, Knoxville
Keywords: parallelization, large systems, massive parallel systems, wide-area networking

PEP: the Phenix Event Processor

Soren P. Sorensen
Geng Han
Qun Li
Qihang Liu
David Morrison
Victor Perevoztchikov
Mark Pollack
Yan Zhao

University of Tennessee

PHENIX Collaboration

The PHENIX detector at the RHIC accelerator will record close to 500 Terabytes
of data annually and will require close to 100 GigaFlops of processing power to
perform the event reconstruction and subsequent data selection procedures. In
order to be able to handle these large requirements the PHENIX Off-line group
has developed a fully distributed event processing application: PEP (The Phenix
Event Processor).

PEP consists of a large set of individual processes performing tasks like:
I/O, access to the central database, display of histograms and events, and
"physics" processing. All these processes can run across a WAN and is
by a single Controler, that keeps track of the state of each process and
allocates resources. The inter-process communication is done via PVM (Parallel
Virtual Machine) from ORNL/Tennessee and all data structures are handled with
the DSPACK data structure package from University of Birmingham. The first
version of PEP is currently implemented on the following UNIX platforms: SGI,
HP, Sun, IBM and Dec. For further information, please refer to: