An Efficient Multiprocessor Management System for Event-Parallel Computing

TL;DR

This paper presents a multiprocessor management system designed to efficiently handle event-parallel computing tasks, demonstrated through a system managing large-scale data reconstruction for high-energy physics experiments.

Contribution

It introduces a new multiprocessor management system optimized for event-parallel computing in high-energy physics data processing environments.

Findings

Improved data reconstruction efficiency for large datasets

Successful management of 8 billion raw events

Enhanced performance in distributed UNIX workstation systems

Abstract

Performance of software using TCP/IP sockets to distribute events to UNIX workstations is described. This simple software was written at the University of Mississippi to control UMiss farm reconstruction of 8 billion raw events, part of Fermilab E791's data. E791 reconstructed HEP's largest data set to study charm physics.