Benes-based Large-capacity Switching Fabric Simulation Study

Recently, to keep pace with continuously rapid development of Internet traffic growth, the technology of terabit or even petabit telecommunication router has become the highlight of researchers of industries and academics. Researchers have been continually exploring new switch architectures. As a core technology of terabit and petabit router, ultra-scalable switch architectures are needed to meet the requirement of capacity, scalability and reliability. How to intelligently implemented a flow control mechanism while maximizing the memory utilization and system performance, how to minimize link congestion and prevent buffer overflow and how to maintain packets' orders if they are delivered over multiple paths are the challenging design issues related to designing the ultra-scalable switch architectures. One method to scale the switch is to use the multi-plane multi-stage buffered architecture. For instance, Cisco's CRS-1 system based on Benes network can scale up to 92 Tbps. This dissertation focuses on the study of scalability, quality of service, fault tolerance, multicast capability, resequencing, and provides several solutions for each issue. This dissertation has shown by simulation that the ultra-scalable multi-plane multi-stage switch architecture is able to perform nearly as well as the output buffered switch under most interested traffic distributions.This dissertation analyzes Cisco's CRS-1 system, describes and studies the properties of the path in Benes network and the non-blocking conditions for Benes network. According to these non-blocking conditions, some algorithms for composing routing labels are discussed.Finding a feasible and efficient load balanced strategy for the ultra-scalable multi-plane multi-stage switch architecture is a top of nowadays research. The dissertation proposes a two-stage load balanced scheme for the ultra-scalable multi-plane multi-stage switch architecture based self-routing and non-blocking permutation benes network. The approach uses reasonable and efficient logical queueing strategy and schedule scheme in ingress traffic managers and switch planes to realize the two-stage load balancing of IP traffic which for different destination addresses.