Physics Simulations

Home
Up
About Me
Projects
Academics
Teaching
Publications
Software
Contests
Personal
Site Map
Overview
Projects
Publications

Overview

Models of Earth's magnetosphere help understand and predict its complex kinetics.  Simulating these physics models are computationally very intensive, and warrant large-scale parallel execution.
  • Traditionally "time-stepped" models have been employed for this purpose.  However, time-stepped methods are severely limited by the "slowest process" in the entire system, even though the required resolution varies by orders of magnitude across different regions.
  • Newer discrete event methods are being developed as a radically different way of approaching this problem[PARA'04].  Parallel execution for these methods presents synchronization and load balancing challenges.

We are developing parallel execution frameworks to (a) enable efficient execution on large distributed memory platforms such as SMP clusters and supercomputers (b) postpone selection of synchronization mechanism all the way up to runtime.

  • The µsik system is our current prototype framework that accommodates a wide range of synchronization mechanisms, both traditional as well as newer ones.  See µsik package webpage.  µsik currently scales to over 128 processors, and supports Time Warp-style optimistic synchronization (with both state saving-based as well as reverse computation-based rollback support), as well as lookahead-based conservative synchronization[Springer'05, PADS'05].  It is now being ported to supercomputing platforms.
  • We are also developing virtualization frameworks to enable the modelers to experiment with different load balancing strategies without having to run actual simulations on supercomputer platforms.
 

Sponsored Projects

Related Publications

  • Parallel Discrete Event Simulations of Grid-based Models -- Asynchronous Electromagnetic Hybrid Code
    Homa Karimabadi, Jonathan Driscoll, Jagrut Dave, Yuri Omelchenko, Kalyan Perumalla, Richard Fujimoto and N. Omidi
    to appear in         Springer Lecture Notes in Computer Science, 2005.
    [Paper]
     
  • A New Approach to Modeling Physical Systems: Discrete Event Simulations of Grid-based Models
    Homa Karimabadi, Yuri Omelchenko, Jonathan Driscoll, N. Omidi, Kalyan Perumalla and Richard Fujimoto
    Workshop on State-of-the-art in Scientific Computing (PARA), June 2004.
    [Paper]
     
  • μsik: A Micro-kernel for Parallel/Distributed Simulation Systems
    Kalyan Perumalla
    to appear in     ACM/IEEE/SCS Workshop on Parallel and Distributed Simulation (PADS), Monterey, CA, June 2005.
    [Paper][Slides]
     
  • Parallel Discrete Event Simulations of Physical Systems using Reverse Computation
    Yarong Tang, Kalyan Perumalla, Richard Fujimoto, Homa Karimabadi, Jonathan Driscoll and Yuri Omelchenko
    to appear in     ACM/IEEE/SCS Workshop on Parallel and Distributed Simulation (PADS), Monterey, CA, June 2005.
    [Paper]
     

[ Back ] [ Home ] [ Up ] [ Next ]
Copyright © Kalyan S. Perumalla.  Last Updated 01/13/2005 10:58 PM -0500.  Disclaimer