A Multilevel Network-assisted Congestion Feedback Mechanism for Network Congestion Control

TL;DR

This paper introduces EECN, a multilevel congestion notification mechanism that improves congestion control efficiency and reduces packet drops, especially benefiting IoT and emerging high-bandwidth applications.

Contribution

It proposes a novel multilevel ECN mechanism and adaptive congestion control responding to network conditions, enhancing existing ECN protocols.

Findings

Packet drop reduced by up to 70%

Flow completion time decreased by 61%

Throughput of short-lived flows improved

Abstract

Network-assisted congestion control leveraging Explicit Congestion Notification (ECN) is an effective way to deal with congestion issues on the Internet. However, we believe that the existing ECN mechanism in the TCP/IP protocol stack may require further optimization to effectively address the evolving congestion challenges introduced by emerging technologies like immersive AR/VR applications and the burgeoning field of the Internet of Things (IoT). To that end, we propose a multilevel congestion notification mechanism called Enhanced ECN (EECN) that leverages the existing two ECN bits in the IP header to notify two levels of congestion in the network and uses the corresponding two bits in the TCP header to negotiate EECN during the handshake and echo congestion experienced back to the sender. Additionally, we propose a congestion control mechanism that triggers different congestion control responses based on the average RTT and multilevel congestion feedback received from the network, which yields promising results, highlighting the effectiveness of utilizing multilevel congestion feedback. The proposed EECN mechanism reduces packet drop by 70% compared to ECN, by 95% compared to TCP New Reno without ECN, and by 40% compared to VCP. The packets marked are reduced by 96% compared to ECN and 76% compared to VCP. Furthermore, the proposed approach reduces flow completion time by 61% compared to ECN and enhances the throughput of short-lived network flows, which are particularly pronounced in IoT environments.