Research On Parallel Processing Of Routing Protocols

Traditional core router, which is the most important infrastructure of Internet, can not meet the multidimensional requirements of next generation Internet , for the limitations of hardware platform, router operating system and routing protocol implementation. Cluster router has more advantages on higher physical ports densities, more scalable forwarding capacities, more powerful computation abilities and lower lifecycle costs. It will be the basis of next generation Internet. With the complicating of router software functions, the super linear increasing of routing information maintained and computed by routing protocols, the requirement of control plane's computation resource exceeds the ability of single processor. It is applied and challenging to research how router software and critical routing protocol modules can utilize distributed computation and storage resources of cluster router to meet the multi-dimension requirements of next generation Internet.The rapid expansion of Internet burdens BGP routing protocol greatly, but traditional BGP implementation is based on single process or single thread and can not scale with the increasing of routing nodes of cluster router. Cluster router distributes routing protocol modules among routing nodes to improve the scalability of its control plane. Each routing protocol module exchanges routing information with adjacent routers and updates local routing node's forwarding table. The key problem of cluster router is to synchronize and maintain the same forwarding tables images among routing nodes.With the application background of massive parallel hybrid cluster router, this dissertation focuses on parallel implemental mechanism of BGP routing protocol and forwarding tables synchronization of cluster router. The main contributions of this dissertation are listed as follows:1. BGP routing protocol's Petri net performance model is proposed by using stochastic reward Petri net (SRN) to analyze the processing path of BGP update packets. The performance model can reflect the performance impacts of different protocol configurations and different packet arriving rates. After the evaluation of BGP performance model under different configuration conditions, it shows that BGP routing protocol is too complicated to support BGP sessions on a large scale and can not meet future Internet requirements.2. Extends traditional BGP software model and proposes a new fully-distributed, highly-parallelized implementation model of BGP and partition algorithm of parallel BGP prefixes computation. The model is composed of BGP entities and C-BGP protocol. BGP entity is the basic function module of prefix computation and storage and can reside on...