Chaotic Routing: An Overview

Of the many router designs, most can be classified as either oblivious or adaptive, depending on whether the path selection is statically determined based on the network topology, or whether dynamic information about congestion, priorities, or faults is considered. Adaptive algorithms can be broken down into two broad categories as well: minimal or non-minimal. Minimal adaptive algorithms allow only topologically minimal-length paths, while non-minimal algorithms allow a larger set of paths, possibly including paths of infinite length. A wide variety of mechanisms exist within each of the categories to provide reliability and performance. Regardless of the method used, the goals of routers are similar. All routers attempt to deliver the greatest number of messages in the least amount of time. Performance parameters for a routing algorithm encompass latency and throughput. Throughput is defined as the rate at which a network delivers data whereas the time for a message to travel from its source to its destination is known as latency. In general for higher performance, latency should be small, and throughput large. Congestion degrades both latency and throughput. Since adaptive routers incorporate congestion information in their routing decisions, they can typically achieve better performance compared to oblivious routing algorithms. Additionally, adaptive schemes inherently have support for building fault tolerant schemes. Minimal adaptive algorithms are simpler to implement but are not as effective as nonminimal adaptive schemes. Non-minimal schemes provide the most flexibility in dealing with congestion but routing decisions are more complex making them harder to implement in hardware. Additional complexity is added because these schemes also need to deal with deadlock, live-lock and out of order packer delivery. In contrast, in oblivious routing algorithm these problems either do no exist or are easy to handle. Despite the improved performance of non-minimal adaptive schemes, they were initially not employed in practical hardware designs due to their complexity. Thus, vast majority of routers continued to use oblivious schemes. Chaotic routing was presented as an alternate non-minimal adaptive scheme, which could be implemented efficiently in hardware unlike its predecessors. With its simpler hardware, it was possible to have a router which performed well under congestion and also enabled good fault tolerance techniques. In this paper we present an overview of the Chaotic routing algorithm, including its operation, performance analysis and fault tolerance techniques. The papers that we have selected echo this. The paper by Bolding, Fulgham and Synder describes the Chaotic routing algorithm and compares its performance with other schemes. The paper by Chinn offers a more detailed analysis of the chaotic router performance under worst case traffic permutations and the paper by Bolding and Yost discusses implementation of fault tolerance schemes on a chaotic router.