STOCHASTIC MODELING FOR HYBRID NETWORK SIMULATIONS

Network-Centric Warfare (NCW) is an objective of the Department of Defense (DoD). By networking large numbers of soldiers, sensors, vehicles, and other equipment, NCW will create an explosion of data in DoD networks. By the same token, Mobile ad hoc networks (MANETs) represent a critical technology for the Armys NCW vision. MANETs are currently an unproven technology that have not been modeled at scales relevant to realistic tactical operations. While research toward understanding the emergent properties of MANETs is progressing, we are far from realizing successful deployment of MANETs even at smaller scales (hundreds of nodes), and this leaves many open questions in realistic settings. Satisfactory evaluation of the technologies supporting these scenarios will not be realized without the development of accurate computer simulation models. These models will involve thousands, perhaps hundreds of thousands of nodes. Such models create prohibitively large state spaces to be accommodated under the current state of the art. Parallelization of network simulation computation offers little benefit due to the end-to-end nature of applications. In addition to the difficulties created by excessively large state spaces, measurements from communication networks indicate that traffic patterns can be bursty and complex [1]. This leads to extremely long simulation time intervals. Standard statistical techniques used in estimating these intervals are not always applicable. To reduce the simulation experiment to a tractable size, researchers seek to incorporate analytical tools into the simulation models. Existing models treat background traffic as deterministic fluids, or as solutions of deterministic differential equations. With these approaches, mixing simulated packet events into the analytically expressed background traffic turns out to be a challenging task. Issues with existing solutions include: i) they miss the incremental impact of newly generated packets into the performance of the overall system, ii) they do not adequately capture stochastic aspects and salient statistical properties of the background traffic, and iii) most importantly, they have degraded accuracy which tends to deteriorate at high traffic loads [2]. To accurately integrate the analytically represented background traffic with the simulated discrete packets, we propose to treat the background traffic as a stochastic process, and introduce a statistically sound method to incorporate the impact of discrete event packets in the overall system performance. Figure 1 shows our method for mixing analytically modeled stochastic background traffic with explicitly generated discrete event packets. Individual packet arrival events are indicated along the bottom axis of the figure. These packets carry with …