Doctor of Philosophy
Department of Electrical, Computer and Telecommunications Engineering
Stacey, Christopher Howard, The distributed server network design problem, Doctor of Philosophy thesis, Department of Electrical, Computer and Telecommunications Engineering, University of Wollongong, 2000. http://ro.uow.edu.au/theses/1360
This thesis considers the design of a Distributed Server Network (DSN). DSN's are required to support the growing number of services where users access resources via a network. Example network services include: the World Wide Web (WWW), distributed database services such as large corporate databases, banking systems, videoconferencing, network games, the Internet's Domain Name System (DNS), Federations of Traders in Open Distributed Processing (ODP) environment, and the X.500 directory service. A DSN comprises an access network connecting clients to servers, and a backbone network interconnecting servers. We call the design of a DSN the Distributed Server Network Design Problem (DSNDP). We examine two closely related problems in the design of DSN's. The first is the design of an entirely new network, given the location of clients, the volume of traffic generated (and required) by each and the possible locations of servers. Here the task is to select: (i) the number, location and capacity of backbone nodes (servers), (ii) the assignment of client nodes to servers, and (iii), the link topology and capacity of both the access and backbone portions of the network. The second form of the problem is the same as the first, except that a link topology is already in place. Here we design a virtual network of data paths and determine the capacity required by them. Three heuristic design procedures for the DSNDP are developed and compared in this dissertation. One procedure is based on a node clustering approach, while the other two are based on a greedy search of the solution space using either ADD/DROP or ADD-A: heuristics. In addition we develop a procedure to provide a lower bound on the cost of a optimal DSN design using a continuous branch-andbound algorithm involving both linear and Sequential Quadratic Programming (SQP).