The performance of distributed simulation depends very much on the partitioning of the simulation model among the participating hosts. Usually, when starting a simulation run, an initial partitioning is determined by taking into account the available computing resources as well as the expected workload and the communication structure of the simulation model. However, as hosts can be subject to background load or the model behavior can change in the course of the simulation, a dynamic partitioning mechanism is required to avoid inefficiencies.In this paper, we introduce a new dynamic partitioning algorithm for optimistic distributed simulation. The algorithm is generally applicable but can also be configured to meet the requirements of specific scenarios. It is based on performance estimates for both computation and communication workload, the calculation of which is completely platform-independent. Our experiments show that the algorithm has low overhead and reacts reliably to changes of both model behavior and external resources.

No abstract

We present a parallel discrete-event simulation method for wireless networks such as mobile ad-hoc networks, mesh networks, or sensor networks. Our method combines new ideas with established approaches and gains considerable speedups over sequential simulations in scenarios commonly used in wireless network simulation studies. In scenarios with only 1000 nodes, the method yields superlinear speedup on up to 6 processor cores without affecting simulation accuracy. A scalability study in scenarios with 2000 nodes demonstrates a speedup of 14 on a multi-core computer with 16 cores. This is good news for a domain generally considered challenging for parallel simulation.

Discrete-event simulation is a very popular technique for the performance evaluation of systems, and in widespread use in network simulation tools. It is well known, however, that discrete-event simulation suffers from the problem of simultaneous events: Different execution orders of events with identical timestamps may lead to different simulation results. Current simulation tools apply tie-breaking mechanisms which order simultaneous events for execution. While this is an accepted solution, a legitimate question is: Why should only a single simulation result be selected, and other possible results be ignored?In this paper, we argue that confidence in simulation results may be increased by analyzing the impact of simultaneous events. We present a branching mechanism which examines different execution orders of simultaneous events, and may be used in conjunction with, or as an alternative to tie-breaking rules. We have developed a new simulation tool, MOOSE, which provides branching mechanisms for both sequential and distributed discrete-event simulation. While MOOSE has originally been developed for network simulation, it is fully usable as a general simulation tool.

No abstract