Tree and ring local access and backbone networks design
Abstract
Computer communication networks (CCNs) have become a vital part of our life in this information age. This thesis deals with the topological design problems of CCNs. CCNs consist of local access networks (LANs) and backbone networks. The LANs support the local traffic to and from the backbone network and the backbone network supports the long haul transmission activities. LANs represent a substantial portion of the investment in CCNs. A problem frequently encountered in LAN design is to form minimum cost ring or tree based networks by deciding how to connect the users to each other and to the backbone nodes. There is a capacity limit imposed by the line capacities or the communication controller port. This thesis attempts to solve the LAN design problem by first forming two matching tours, then dividing them into clusters that satisfy the capacity constraints by using an optimal partitioning algorithm, and finally forming small tours for the ring network and minimum spanning trees for the tree network. The computational results are compared with other algorithms. One of the main problems faced by the backbone network designer is to decide the traffic routing, the links to be included in the design and their capacities to minimize the network cost with the response time constraints. This thesis tackles the problem by finding the routing and the links simultaneously in order to minimize the delay cost, the fixed link set up cost, and the variable link utilization cost of a network. This thesis presents an approach to design a overall network. It starts with the tree configuration of the local access networks and then uses the outputs of the tree network design to the backbone network design. The problems mentioned above are formulated as combinatorial optimization problems, that as it turns out, belong to NP-Complete class. They are unlikely to have optimal solution procedures in reasonable times. Hence, heuristics are developed for both kinds of local area networks and the backbone network. Preliminary tests show that the 2-matching heuristics have better results (average 38.4% better for tree and 39.6% better for ring) than the Parallel Savings Algorithms in the multicenter cases. The new heuristics also have worst case error guarantees which are 3 for the tree networks and 2 for the ring networks when the sizes of all 2-matching tours are greater than K.
Degree
Ph.D.
Advisors
Altinkemer, Purdue University.
Subject Area
Management|Computer science
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.