Parallel discrete event simulations of grid-based models: asynchronous electromagnetic hybrid code

  • Authors:

  • Homa Karimabadi;Jonathan Driscoll;Jagrut Dave;Yuri Omelchenko;Kalyan Perumalla;Richard Fujimoto;Nick Omidi

  • Affiliations:

  • SciberNet, Inc, Solana Beach, CA;SciberNet, Inc, Solana Beach, CA;,SciberNet, Inc, Solana Beach, CA;SciberNet, Inc, Solana Beach, CA;Georgia Institute of Technology, Atlanta, GA;Georgia Institute of Technology, Atlanta, GA;SciberNet, Inc, Solana Beach, CA

  • Venue:
  • PARA'04 Proceedings of the 7th international conference on Applied Parallel Computing: state of the Art in Scientific Computing
  • Year:
  • 2004

Quantified Score

Hi-index 0.00

Visualization

Abstract

The traditional technique to simulate physical systems modeled by partial differential equations is by means of a time-stepped methodology where the state of the system is updated at regular discrete time intervals. This method has inherent inefficiencies. Recently, we proposed [1] a new asynchronous formulation based on a discrete-event-driven (as opposed to time-driven) approach, where the state of the simulation is updated on a “need-to-be-done-only” basis. Using a serial electrostatic implementation, we obtained more than two orders of magnitude speedup compared with traditional techniques. Here we examine issues related to the parallel extension of this technique and discuss several different parallel strategies. In particular, we present in some detail a newly developed discrete-event based parallel electromagnetic hybrid code and its performance using conservative synchronization on a cluster computer. These initial performance results are encouraging in that they demonstrate very good parallel speedup for large-scale simulation computations containing tens of thousands of cells, though overheads for inter-processor communication remain a challenge for smaller computations.