Quantitative assessment of an agent-based simulation on a time warp executive

Abstract

We recently introduced SASSY, the design for a hybrid simulator that provides an agent-based API atop a PDES kernel (Hybinette et al. 2006). Our hypothesis is that a design like SASSY offers the advantages of an agent-based paradigm for the application developer, but also provides the performance advantages of a PDES kernel. Since the time of our initial publication, most aspects of SASSY's design have been implemented, and we are now assessing our hypotheses e.g., (He and Hybinette 2008). In this paper we investigate performance advantages for a simple agent-based application on SASSY. In most cases, agentbased simulation environments are configured using a time-step approach, where the simulation proceeds in discrete steps. In this paper we evaluate the performance of a simple application running in a traditional time-step simulation, and also its performance when running on SASSY with PDES support.

On-the-fly parallelization in agent-based simulation systems

“…Since each SASSY Agent runs in its own thread and needs access to the world state, there is one copy of the state given to each Agent to prevent the need for unnecessary locks that would slow down the system. We are aware of this memory problem and have studied solutions to it in previous work (Vulov et al 2008).…”

“…This API is based on a sense-think-act cycle and logically separates an agent's brain from its body (Hybinette et al 2006). Experiments in (Vulov et al 2008) show that SASSY can achieve significant speedup (6.2x) when run on 8 processing elements (PEs) with thousands of agents.…”

Collaborative Interest Management for Peer-to-Peer Networked Virtual Environment

“…Some of them apply similar ideas of interest management used in agent-based simulation, that is, some specific processes are created to perform interest management in the middleware [23], [24], [25]. Instead of using specific processes in the middleware, the existing P2P architectures to support NVEs usually select some peers to perform either traditional cell-based IM mechanism or area-based IM mechanism [15], [17], [34].…”

“…Two common approaches to reduce the status updates between players are dead reckoning [19], [20] and Interest Management (IM) [21], [22]. In addition to NVE, IM techniques have been widely adopted in many other research fields such as agent-based simulation [23], [24], [25] and most recently, wireless sensor network [26]. It utilizes filtering mechanism to avoid broadcasting communication among processes or network nodes.…”