Adaptive Cheapest Path First Scheduling in a Transport-Layer Multi-Path Tunnel Context

TL;DR

This paper introduces the Adaptive Cheapest Path First (ACPF) scheduler, which improves multi-path network performance by dynamically adjusting path usage to prevent congestion, outperforming the traditional CPF approach.

Contribution

The paper proposes and evaluates the ACPF scheduler that adapts path selection based on congestion, addressing limitations of the CPF scheduler in multi-path transport layers.

Findings

ACPF improves average throughput by 24% to 86%.

ACPF reduces queue buildup and congestion.

The scheduler is effective over TCP and DCCP tunnels.

Abstract

Bundling multiple access technologies increases capacity, resiliency and robustness of network connections. Multi-access is currently being standardized in the ATSSS framework in 3GPP, supporting different access bundling strategies. Within ATSSS, a multipath scheduler needs to decide which path to use for each user packet based on path characteristics. The Cheapest Path First (CPF) scheduler aims to utilize the cheapest path (e.g. WiFi) before sending packets over other paths (e.g. cellular). In this paper, we demonstrate that using CPF with an MP-DCCP tunnel may lead to sub-optimal performance. This is due to adverse interactions between the scheduler and end-to-end and tunnel congestion control. Hence, we design the Adaptive Cheapest Path First (ACPF) scheduler that limits queue buildup in the primary bottleneck and moves traffic to the secondary path earlier. We implement ACPF over both TCP and DCCP congestion controlled tunnels. Our evaluation shows that ACPF improves the average throughput over CPF between 24% to 86%.