Achieving Low Latency and High Throughput over Cellular Internet Connections

Abstract: The continuous increase in the number and type of Internet services and their requirements for improved QoS has motivated the steady evolution  of cellular networks towards the current fifth generation (5G) systems. However, updating the network to 5G is insufficient to satisfy application requirements since performance limitations can also exist in the transport used by the applications. Legacy transport protocols and congestion control algorithms (CCAs) are not suitable for applications with requirements for both throughput and delay. This mismatch has inspired new transport protocols and CCAs, such as QUIC and BBR. Nevertheless, cellular networks present challenges that can make it difficult for newly proposed CCAs to achieve consistent throughput and delay.       The main focus of this thesis is enhancing transport protocols and CCAs to achieve lower delay and high throughput in cellular networks. An extensive review of available end-to-end CCAs for cellular networks is provided in this thesis, along with the challenges and future directions for research. The delivery rate at a receiver is an important quantity that has found increased use in modern CCAs, and in this thesis, we propose and validate a Kalman-filter-based technique to obtain a steady estimate of the delivery rate for a cellular bottleneck. This thesis also proposes an extension to the QUIC protocol to make receiver-side delivery rate estimates available to the sender CCAs. Using the proposed rate estimation method and extension to the QUIC protocol, this thesis proposes modifications to the recently proposed CCAs BBR and Copa. The proposed modifications are evaluated over real cellular networks and through extensive trace-based emulations. The modified BBR results in lower packet delays with similar throughput to standard BBR in cellular bottlenecks. On the other hand, the modification to Copa strives to provide a more consistent and predictable delay performance across different cellular bottlenecks.