Fair Resource Sharing for Stateless-Core Packet-Switched Networks With Prioritization

We propose an activity-based congestion management (ABC) to enforce fair bandwidth sharing among users in packet-based communication networks without requiring per-user information in forwarding nodes. Activity relates to the sent data rate of a user and its contracted reference rate. Activity meters monitor user traffic at edge nodes and add activity information to packets. Forwarding nodes utilize this information to preferentially drop packets with high activity values in case of congestion. In this paper, we significantly simplify the behavior of ABC's control devices: the activity meter in edge nodes and the activity AQM in forwarding nodes. Moreover, we introduce scheduling priorities for ABC, provide means to avoid starvation of non-prioritized traffic through prioritized traffic, and means to disincentivize prioritized transmission of non-realtime traffic. We study the fairness achieved with and without ABC using packet-based simulation, mostly under challenging conditions where a heavy user wants to monopolize a bottleneck's bandwidth. We investigate bandwidth sharing for non-responsive traffic, for responsive traffic, and a mixture of both. The effect of traffic prioritization is studied. We explore the impact of configuration parameters and recommend meaningful system parameters. We illustrate their impact on system dynamics and show that ABC expedites sporadic uploads, which improves the quality of experience for interactive applications. Finally, we compare the performance of ABC with the one of CSFQ whose objective is the same as the one of ABC, but significantly differs with regard to signaled information and algorithmic approach. Summarizing, ABC creates an ecosystem, where users can maximize their throughput if they do not exceed their fair share on the bottleneck link. This incentivizes the usage of congestion-controlled transport protocols. Moreover, ABC protects traffic from users applying congestion control against traffic overload from users not leveraging congestion control, even if the latter send prioritized traffic.