Scalable RTI-Based Parallel Simulation of Networks

  • Authors:
  • Kalyan S. Perumalla;Alfred Park;Richard M. Fujimoto;George F. Riley

  • Affiliations:
  • College of Computing, Georgia Tech, Atlanta, Georgia;College of Computing, Georgia Tech, Atlanta, Georgia;College of Computing, Georgia Tech, Atlanta, Georgia;Department of Electrical and Computer, Engineering, Georgia Tech, Atlanta, Georgia

  • Venue:
  • Proceedings of the seventeenth workshop on Parallel and distributed simulation
  • Year:
  • 2003

Quantified Score

Hi-index 0.00

Visualization

Abstract

Federated simulation interfaces such as the High LevelArchitecture (HLA) were designed for interoperability,and as such are not traditionally associated with highperformancecomputing. In this paper, we present resultsof a case study examining the use of federated simulationsusing runtime infrastructure (RTI) software to realizelarge-scale parallel network simulators. We examine theperformance of two different federated networksimulators, and describe RTI performance optimizationsthat were used to achieve efficient execution. We showthat RTI-based parallel simulations can scale extremelywell and achieve very high speedup. Our experimentsyielded more than 80-fold scaled speedup in simulatinglarge TCP/IP networks, demonstrating performance of upto 6 million simulated packet transmissions per second ona Linux cluster. Networks containing up to two millionnetwork nodes (routers and end systems) were simulated.