Fully-Distributed and Highly-Parallelized Implementation Model of BGP4 Based on Clustered Routers

摘要

With the explosive growth of the service providers topologies, the size of routing table and number of routing sessions have experienced a sharp increase, which makes the scaling of BGP implementation in backbone routers a rather stressing problem. Clustered router is a promising architecture in terms of forwarding capability. But traditional BGP implementation, based on single-processor architecture cannot utilize the distributed computing and memory resources to enhance the protocol's performance. Borrowing the idea of team working from Multi-Agent technology, this paper proposes a new fully-distributed and highly-parallelized implementation model of BGP, which gives equal chance to each routing node of the cluster to participate in routing selection and stores route entries among routing nodes. Chief algorithms are presented related with session dispatch and partition of prefixes computing and storage. Theory analysis and experiment show that the algorithm can achieve linear speedup over central control model and increase system reliability greatly.