Ultra Large-Scale Wireless Network Models using Massively Parallel Discrete-Event Simulation

摘要

Mobile ad-hoc networks (MANETs) are becoming pervasive in our always-connected lives of iPhones, laptops, etc. While entertainment is important, next generation military applications can and will come to depend on these new technologies. Connectivity can no longer be optional; it has become a must-have requirement. Over the past decade, the DoD has spent billions of dollars investigating new approaches to radio including the Joint Tactical Radio System (JTRS) and Software-Defined Radios (SDRs). However, large-scale deployments (1 million+ nodes) have never been tested for a single homogeneous network of that magnitude. The goal of this paper is to demonstrate the efficacy of large-scale wireless network simulations on multiple High-Performance Computing (HPC) platforms. By exploiting the strengths of two different supercomputers, we evaluate our million node Optimized Link State Routing (OLSR) network on the CCNI Blue Gene/L as well as the Army Research Lab's SGI ICE 8200 (Harold) compute cluster. By exploiting infrequent synchronization we may push the simulation to its maximum potential for our model. We realize speedups between 315-450 in our normalized parallel simulation over sequential.