Compact Route Computation: Improving Parallel BGP Route Processing for Scalable Routers

Nowadays Internet routers are overwhelmed by a large quantity of BGP (Border Gateway Protocol) updates triggered by route changes. The fast growth of the Internet size further aggravates router workloads and exacerbates routing convergence performance. Scalable routers, such as cluster routers and ForCES, are proposed to exploit distributed control plane (DCP) with multiple control elements (CEs) to scale route processing capacity. Previous studies show that most route updates are duplicated in the Internet. Traditional parallel computation schemes only consider calculating route in parallel but most routes are computed in vain since they will not be finally selected. This paper proposes a simple and novel idea of compact route computation (CRC) to reduce BGP route processing load and improve routing convergence performance. Our scheme partitions Adj-RIBs-in among multiple CEs in the granularity of prefixes, which makes non-consecutive updates for a prefix queued adjacently in distributed control plane. Route computations triggered by a prefix's consecutive updates are compacted into one. We evaluate our scheme by simulations with real BGP update data, and results show that our scheme is very effective to reduce route computation workloads. For example, for a scalable router with 4 CEs and the updates received from 24 neighbors, our scheme reduces 60% route computation load. It can distinctly reduce route computation load with more CEs.