Multiple ring networks in clustered traffic environments.

Abstract

Ring networks are appropriate for the full range of network levels, including multiprocessor systems, local area computer networks and high speed backbones. The most well known and widely implemented examples are the IBM token ring and FDDI networks. Ring networks have the advantages of high channel utilisation and bounded delay if an n-limited service policy is used. The packet transfer delay, defined as the average time a packet spends in the network from the time it is generated until the time it is received at its destination node, improves with the number of rings on which a node is connected. However, many ring connections are not economically feasible since the cost of the ring interface increases with the number of rings. There has been an abundance of previous work on single token ring networks. A number of papers on slotted rings, register insertion rings and more complex ring architectures have also been published. However, there is very little existing literature on multiple ring networks as well as ring networks in clustered traffic environments, i.e. where nodes from the same cluster tend to communicate more with each other than with other nodes in the network. This thesis focuses on two network topologies that make use of multiple rings and are well suited to clustered traffic environments: the two-connected multiple ring (2-MR) and the destination removal double ring (DRDR). For the 2-MR network, three different practical token-based protocols are investigated in an attempt to optimise performance. It is further shown that significant performance improvements can be achieved by employing a slotted ring protocol rather than the token ring protocol. The DRDR network is also examined and its performance compared to the aforementioned architectures. For each of the six cases, both random and clustered traffic patterns are considered and compared. Analytical results are derived which are verified by results obtained from computer simulations. Furthermore, we look at exact methods of analysing ring networks. A mean value analysis of a single token ring network with a I-limited service discipline is performed, which clearly shows the complexity exact methods introduce. Finally, although it has been stated in the literature that an exact analysis of a multiple symmetrical token ring network is intractable, we present a novel Markov chain approach that gives exact results for near zero loads.