A Scalable Multithreaded BGP Architecture for Next Generation Router

To satisfy the requirement for the large-scale and high-speed Internet, the critical interdomain routing standard BGP on core routers must address the challenges of performance and scalability. The introduction of multicore and multithreading technique could effectively resolve the computing bottleneck in singlecore processor and the view consistency problem in cluster system, thus becoming an important approach to further improve the performance of core routers. In this paper, a multithreaded BGP architecture named Threaded BGP (TBGP) is proposed based on neighbor session division, achieving the parallel processing of different sessions on multiple threads by employing data parallelism. Experimental results show that TBGP yields good performance improvement compared to traditional BGP, where the average speedups of route learning time for TBGP under iBGP and eBGP sessions reach 1.79 and 2.73, and its peer switch time is decreased greatly by 45% on average with session number increasing to 100, 200 and 300. Meanwhile, TBGP scales better than BGP by improving the CPU efficiency so that it processes more routes when affording 300 peer sessions under iBGP and eBGP sessions.