Techniques for large scale distributed simulations of computer networks.

摘要

Discrete event simulation is widely used within the networking research community to demonstrate the validity of various network protocols and architectures. Many of the current advances in networking technology were first demonstrated and tested using simulation. The construction of a discrete event simulator to model a computer network is a simple and straightforward activity. One needs a method of describing the topology of the network to be modeled, a method for describing the behavior of the network elements, a simulation engine that manages and processes the pending events queue, and a method for observing the behavior of those elements. However, when constructing a simulator in this fashion one quickly runs into difficulties. The simulation of any reasonable size network (of more than a few hundred high speed network elements) can take excessive amounts of computer resources, both in CPU time and memory.; This thesis seeks to overcome these difficulties by defining methods to run a single network simulation in a parallel and distributed environment. Using a parallel and distributed simulation environment, the overall network model is simulated on a number of workstations connected by some interconnect hardware. Each workstation is responsible for modeling events of only a portion of the overall network being simulated. The workstations exchange information via the interconnect when events generated on one workstation affect network elements modeled on another workstation. With this distributed technique, the overall size of the network being modeled can grow almost linearly with the number of workstations. Additionally, the overall execution time can be reduced due to parallel processing of events.