Scalable RTI-based parallel simulation of networks

Abstract

Federated simulation interfaces such as the High Level Architecture (HLA) were designed for interoperability, and as such are not traditionally associated with highperformance computing. In this paper, we present results of a case study examining the use of federated simulations using runtime infrastructure (RTI) software to realize large-scale parallel network simulators. We examine the performance of two different federated network simulators, and describe RTI performance optimizations that were used to achieve efficient execution. We show that RTI-based parallel simulations can scale extremely well and achieve very high speedup. Our experiments yielded more than 80-fold scaled speedup in simulating large TCP/IP networks, demonstrating performance of up to 6 million simulated packet transmissions per second on a Linux cluster. Networks containing up to two million network nodes (routers and end systems) were simulated.

Performance of a mixed shared/distributed memory parallel network simulator

“…With an increased local to message-passing communication ratio, message-passing has less impact on performance resulting in the gap between the performance of the Mixed and Message-Passing communication models decreasing. For similar NMS baseline benchmarks performed using DaSSFNet [9], GTNetS and pdns [7,13], these performance results, in terms of the level of packet transmissions per second achieved, are some of the best for the number of processors used. Note that the benchmarks for the different simulators were performed on different computer systems making it difficult to give exact comparisons.…”

“…Examples of this approach are given in [7,13] for the pdns [16] and GTNetS [14] network simulators. A runtime infrastructure (RTI) software package called RTI-Kit [6] provides communication and synchronization between federates.…”

Large-scale network simulation: how big? how fast?

“…At the heart of TM-Kit is the computation of reductions (global minimums) that account for transient messages using message counters. This computation and performance optimizations that have been developed are described in [29].…”

µsik — A Micro-Kernel for Parallel/Distributed Simulation Systems

“…Their combined performance can be expected to be better than each of them running individually and separately. Our current implementation includes two different modules: one is based on efficient global hierarchical reductions [20,22], while the other is based on an optimized variant [23] of the Chandy-Misra-Bryant null message algorithm [24]. These have been tested on large-scale platforms, and have been demonstrated to scale very well, even up to supercomputing configurations of more than 1500 processors [22,23,25].…”