可扩展路由器并行路由计算模型

分布式控制平面的并行路由计算性能是制约可扩展路由器大规模扩展的关键因素。根据反压理论建立了分布式控制平面并行路由计算模型,利用控制单元的队列长度信息判断每个控制单元的负载大小,自适应地调整控制单元间的负载分配,实现负载均衡。利用排队理论,根据损失率、系统利用率和服务等待时间这3个特征量对该模型进行了性能分析。用真实网络数据模拟验证理论分析的正确性。实验结果表明:与现有并行路由算法相比,该模型采用的反压任务分配算法能够根据控制单元的负载大小和控制单元数量自适应地均衡分配负载,有效地提高了可扩展路由器的并行路由计算性能和可扩展性。

Parallel route computing model for scalable routers

The performance of parallel route computations in distributed control planes is key to high scalability of scalable routers.This paper presents a parallel route computation model for distributed control planes derived from the back-pressure theory.A back-pressure task dispatching scheme is used to achieve load balance with route computations tasks divided among different control elements according to the load information of each control element.The system performance is evaluated based on loss probability,server utilization and expected response time.Tests with real network data show that the back-pressure task dispatching scheme effectively improves the performance of parallel route computations and the scalability of scalable routers compared with traditional parallel route computation schemes.