Pelegant : a parallel accelerator simulation code for electron generation and tracking.
Abstract
elegant is a general-purpose code for electron accelerator simulation that has a worldwide user base. Recently, many of the time-intensive elements were parallelized using MPI. Development has used modest Linux clusters and the BlueGene/L supercomputer at Argonne National Laboratory. This has provided very good performance for some practical simulations, such as multiparticle tracking with synchrotron radiation and emittance blow-up in the vertical rf kick scheme. The effort began with development of a concept that allowed for gradual parallelization of the code, using the existing beamline-element classification table in elegant. This was crucial as it allowed parallelization without major changes in code structure and without major conflicts with the ongoing evolution of elegant. Because of rounding error and finite machine precision, validating a parallel program against a uniprocessor program with the requirement of bitwise identical results is notoriously difficult. We will report validating simulation results of parallel elegant against those of serial elegant by applying Kahan's algorithm to improve accuracy dramatically for both versions. The quality of random numbers in a parallel implementation is very important for some simulations. Some practical experience with generating parallel random numbers by offsetting the seed of each random sequence according to the processor ID willmore »
- Authors:
-
- (APS)
- Publication Date:
- Research Org.:
- Argonne National Lab. (ANL), Argonne, IL (United States)
- Sponsoring Org.:
- USDOE
- OSTI Identifier:
- 973755
- Report Number(s):
- ANL/AES/CP-119207
TRN: US1002023
- DOE Contract Number:
- DE-AC02-06CH11357
- Resource Type:
- Conference
- Resource Relation:
- Conference: Advanced Accelerator Concepts (AAC06); Jul. 10, 2006 - Jul. 15, 2006; Lake Geneva, WI
- Country of Publication:
- United States
- Language:
- ENGLISH
- Subject:
- 43 PARTICLE ACCELERATORS; ACCELERATORS; ACCURACY; ALGORITHMS; ANL; CLASSIFICATION; ELECTRONS; IMPLEMENTATION; PERFORMANCE; SIMULATION; SUPERCOMPUTERS; SYNCHROTRON RADIATION
Citation Formats
Wang, Y., Borland, M. D., and Accelerator Systems Division. Pelegant : a parallel accelerator simulation code for electron generation and tracking.. United States: N. p., 2006.
Web. doi:10.1063/1.2409141.
Wang, Y., Borland, M. D., & Accelerator Systems Division. Pelegant : a parallel accelerator simulation code for electron generation and tracking.. United States. doi:10.1063/1.2409141.
Wang, Y., Borland, M. D., and Accelerator Systems Division. Sun .
"Pelegant : a parallel accelerator simulation code for electron generation and tracking.". United States.
doi:10.1063/1.2409141.
@article{osti_973755,
title = {Pelegant : a parallel accelerator simulation code for electron generation and tracking.},
author = {Wang, Y. and Borland, M. D. and Accelerator Systems Division},
abstractNote = {elegant is a general-purpose code for electron accelerator simulation that has a worldwide user base. Recently, many of the time-intensive elements were parallelized using MPI. Development has used modest Linux clusters and the BlueGene/L supercomputer at Argonne National Laboratory. This has provided very good performance for some practical simulations, such as multiparticle tracking with synchrotron radiation and emittance blow-up in the vertical rf kick scheme. The effort began with development of a concept that allowed for gradual parallelization of the code, using the existing beamline-element classification table in elegant. This was crucial as it allowed parallelization without major changes in code structure and without major conflicts with the ongoing evolution of elegant. Because of rounding error and finite machine precision, validating a parallel program against a uniprocessor program with the requirement of bitwise identical results is notoriously difficult. We will report validating simulation results of parallel elegant against those of serial elegant by applying Kahan's algorithm to improve accuracy dramatically for both versions. The quality of random numbers in a parallel implementation is very important for some simulations. Some practical experience with generating parallel random numbers by offsetting the seed of each random sequence according to the processor ID will be reported.},
doi = {10.1063/1.2409141},
journal = {},
number = ,
volume = ,
place = {United States},
year = {Sun Jan 01 00:00:00 EST 2006},
month = {Sun Jan 01 00:00:00 EST 2006}
}