MR-PBGP: A Multi-Root Tree Model for Parallel BGP

BGP protocol has been confronted with many performance problems caused by large scale networks, and have raised great concerns of industry and academia. In this paper, we present a multi-root tree model called MR-PBGP for parallel BGP in the context of a multi-core platform with shared memory. This model is an integrated model of the neighbor-based division and data division. We prove that threads of BGP constructed by a multi-root tree with binary bough excluding the first-level branch nearest leaves will achieve optimal performance with unlimited number of cores (ideal case) in a router. However, with limited number of cores (actual case), we proposed an approximation algorithm to solve this nonlinear programming problem. Simulation results show that (i) the approximation algorithm can get a good performance speedup and parallel efficiency w.r.t. the increase of its neighbors. (ii) 50% of the optimal performance can be reached by using only 5% of the ideal number of cores. Comparing with other researches, MR-PBGP is a more effective model for parallel BGP.