APIM: Adaptive Path Isolation for Mixed Flows in Data Center Networks

With the emergence of 5G and cloud computing technologies, modern data centers host a wide variety of heterogeneous mixed flows. However, only few load balancing schemes adopt path partition to route mixed flows. Due to resource competition between mouse and elephant flows, two key performance indicators, flow completion time and throughput, are negatively impacted to some extent. We propose a new load balancing scheme APIM. Based on adaptive path isolation, the elephant and mouse flows are forwarded on their respective paths. Meanwhile, APIM uses Equal Cost Multi Path for mouse flows to acquire low latency and Random Packet Spraying for elephant flows to guarantee throughput. The experimental results show that APIM decreases the completion time of mouse flows and ensures the throughput of elephant flows.