Architectures and Algorithms for Transport-layer Multi-connectivity in Next Generation Wireless Networks

Abstract

In this thesis, we make four contributions to Transport-layer Multi-connectivity (TLMC) in wireless networks. First, we address the challenges in deploying and using MPTCP as a higher-layer steering function in the 5G Access Steering Splitting Switching (ATSSS) function [89]. We observe that while the design choices of MPTCP have given it broad support for existing applications, they did not address the issue of deploying it in devices with proprietary kernels, such as the majority of smart devices. Second, we address the problem of scheduling packet transmissions amongst multiple wireless paths with uncertain, time-varying delay. We make the observation that the requirement for multi-path scheduling is usually to transmit application layer objects (web pages, images, video frames etc) with low latency, and so it is the object delay rather than the per packet delay which is important. This has fundamental implications for multipath scheduler design. We introduce SOS (Stochastic Object- aware Scheduler), a multipath scheduler that considers application layer object sizes and their relationship to link uncertainty. Third, we analyze the challenges faced by MPTCP when used to aggregate multiple WAN/Internet connections in Multi-WAN Routers (MWR). We analyze two different architectural variants of MPTCP in MWR and show that they suffer from performance issues. Instead, we propose a new multi-path solution more suited to MWR, called BOOST, which eliminates issues with MPTCP in MWR. Finally, we consider the task of transporting mission-critical Train to Ground (T2G) traffic, such as Closed-Circuit TV (CCTV) and Communication-Based Train Control (CBTC), which have strict QoS requirements for high availability and low packet loss, in underground trains that are equipped with multiple WiFi backhauls. Trains suffer from frequent handovers that significantly impact real-time mission-critical applications. we take a machine-learning approach of predicting when a handover is about to occur, using features extracted only from WiFI logs, and replicating packets shortly before they happen, using minimal redundancy while still significantly reducing handover losses.