Data classification and reinforcement learning to avoid congestion on SDN-based data centers
Gustavo Diel, Charles C. Miers, Maurício A. Pillon, Guilherme Koslovski · GLOBECOM 2022 - 2022 IEEE Global Communications Conference · 2022
A contemporaneous data center (DC) hosts multiple competitive network data flows from different applications, sharing the intermediate switches capacities. In this context, congestion control and avoidance on Transmission Control Protocol (TCP) are critical tasks to ensure the quality of service for hosted applications. Specifically, Software Defined Networking (SDN) created an opportunity to avoid congestion once the centralized controller can gather ongoing and historical information from all network switches and flows. However, the data gathered is enormous, and fast-computing algorithms are crucial for decision-making. In this sense, this work proposes Reinforcement Learning- and SDN-aided Congestion Avoidance Tool (RSCAT), which uses data classification to determine if the network is congested and actor-critic reinforcement learning to find better TCP parameters. Our experimental analysis shows RSCAT could decrease the Flow Completion Time (FCT) of DCTCP and CUBIC variants in several cases without requiring any software update on DC end-points.