Modelling adaptive routing in Wide Area Networks

Hutchison, Andrew (1991)

Bibliography: leaves 132-138.

Thesis

This study investigates the modelling of adative routing algorithms with specific reference to the algorithm of an existing Wide Area Network (WAN). Packets in the network are routed at each node on the basis of routing tables which contain internal and external delays for each route from the node. The internal delay on a route represents the time that packets queued for transmission will have to wait before being transmitted, while the external delay on a route represents the delay to other nodes via that route. Several modelling methods are investigated and compared for the purpose of identifying the most appropriate and applicable technique. A model of routing in the WAN using an analytic technique is described. The hypothesis of this study is that dynamic routing can be modelled as a sequence of models exhibiting fixed routing. The modelling rationale is that a series of analytic models is run and solved. The routing algorithm of the WAN studied is such that, if viewed at any time instant, the network is one with static routing and no buffer overflow. This characteristic, together with a real time modelling requirement, influences the modelling technique which is applied. Each model represents a routing update interval and a multiclass open queueing network is used to solve the model during a particular interval. Descriptions of the design and implementation of X wan, an X Window based modelling system, are provided. A feature of the modelling system is that it provides a Graphical User Interface (GUI), allowing interactive network specification and the direct observation of network routing through the medium of this interface. Various applications of the modelling system are presented, and overall network behaviour is examined. Experimentation with the routing algorithm is conducted, and (tentative) recommendations are made on ways in which network performance could be improved. A different routing algorithm is also implemented, for the purpose of comparison and to demonstrate the ease with which this can be affected.

Collections: