Introduction

Pelegant stands for ``parallel elegant'', which is a parallelized version of elegant [1]. Written in the C programming language with MPICH, the Pelegant has been successfully ported to several clusters and supercomputers, such as the ``weed'' cluster at APS, the Fusion cluster at Argonne National Lab, and the BlueGene supercomputer at Argonne Leadership Computing Facility. Thanks to careful design in parallelization and good architecture of the serial elegant, the Pelegant achieves very good performance. For example, for a simulation of $100,000$ particles in APS including symplectic element-by-element tracking, accelerating cavities, and crab cavities, the simulation time was reduced from 14.3 days to 42 minutes on 512 CPUs of the BlueGene/L (BG/L) supercomputer. The speedup for this particular simulation is 484 with efficiency near 95%.

This document describes how to build Pelegant, run the code and optimize the performance. An explicit list of elements which have NOT been parallelized can be found in the appendices. The user can also check if an element has been parallelized from the latest elegant manual. The list of commands which have been used in the regression tests is available at the end of the appendices. The user should be familiar with the User's Manual for elegant before reading this document.