DRL-PLink: Deep Reinforcement Learning With Private Link Approach for Mix-Flow Scheduling in Software-Defined Data-Center Networks

In datacenter networks, bandwidth-demanding elephant flows without deadline and delay-sensitive mice with strict coexist. They compete each other for limited network resources, the effective scheduling of such mix-flows is extremely challenging. We propose a deep reinforcement learning private link approach (DRL-PLink), which combines software-defined (DRL) to schedule mix-flows. DRL-PLink divides bandwidth establishes some corresponding private-links different types isolate them that competition among can decrease accordingly. DRL used adaptively intelligently allocate resources these private-links. Furthermore, improve policy, introduces novel clipped double Q-learning, exploration noise, prioritized experience replay technology DDPG address function approximation error, induce lager more randomness exploration, as well efficient in respectively. The experiment results under actual workloads (including Web search data mining workload) indicate effectively at small system overhead. Compared ECMP, pFabric, Karuna, average flow completion time decreased by 77.79%, 65.61%, 23.34% respectively, when meet rate increased 16.27%, 0.02%, 0.836% Additionally, also achieve load balance between paths.