基于SDN的混合分段路由概率流调度机制

针对数据中心网络流量路径分配不均匀、易造成大流碰撞，以及控制器流表开销大等问题，提出了一种基于SDN的混合分段路由概率流调度机制SRPFS（segment routing probability flow scheduling）。利用SDN集中控制与全局视图特性，首先采用混合分段路由完成流量初始转发；然后选用粒子群优化算法，重定义粒子群内部寻优过程来对流量进行筛选；最后构造全局节点概率矩阵，设计概率调度算法选举出流量转发最优路径。实验结果表明混合分段路由转发技术在流表开销方面优势较大，并且SRPFS相比于其他较典型的流传输机制，在平均网络吞吐量、链路利用率、标准网络吞吐率等方面有明显优势，能够有效减轻控制器的流表负载，保证了较好的网络性能。

SDN-based hybrid segmented routing probabilistic flow scheduling mechanism

In order to solve the problems of unreasonable traffic path allocation, elephant flow collision and high overhead of controller flow table in data center networks, this paper proposed an SDN based hybrid segment routing probability flow scheduling mechanism(SRPFS). By taking advantage of SDN''s centralized control feature, SRPFS initialized the traffic forwarding method by hybrid segment routing at first. Then, it utilized particle swarm optimization and redefined the internal particle''s optimization process to filter the traffic. Finally, SRPFS constructed the global node probability matrix and designed probability scheduling algorithm to elect the optimal path for traffic forwarding. Experimental results show that the hybrid segment routing has a greater advantage in terms of flow table overhead. Compare with other classic flow transmission mechanism, SRPFS has certain advantages in average network throughput, link utilization and standard network throughput rate, it reduces the controller flow-table overhead and ensures better network performance.