By Rahil Gandotra, CableLabs; Arun Yerra, CableLabs
Future networks (fixed and mobile) are gearing up for demanding applications like immersive XR, selfdriving cars, and healthcare robots. These applications are expected to demand more from the network in terms of QoS characteristics. In particular, such applications require low latency/jitter, high data rates, and highly reliable and available networks. Packets not delivered within the required latency/jitter budget will be wasted and the user experience will be significantly impacted.
The transport layer, operating between the network and application layers, is the first layer in the stack that functions on an end-to-end basis between the two communicating hosts. User experience and overall network performance depend heavily on how applications, the transport layer, and the network work in synergy. Transport protocols provide several critical functions to enable data exchange between applications on a network: process-to-process delivery, multiplexing and demultiplexing, flow control, congestion control, etc. The increasing heterogeneity of the network deployment scenarios and the diverse and challenging QoS requirements make the role of transport protocols more crucial and more complex to design.
The adoption of new transport-layer solutions is restricted due to several factors, and the research community is forced to work around these limitations and design innovative approaches to improve network performance. The widespread use of middleboxes, which often block unknown protocols or unrecognized extensions to known protocols, invalidates the end-to-end principle, thereby impeding the deployment of alternative protocols, leading to transport protocols ossification [1]. Furthermore, most operating systems implement transport functionalities (e.g., TCP and UDP) within the kernel space, exposing socket APIs to the applications, making the deployment of new solutions difficult and limiting the interfacing options between applications and the transport protocols. This has essentially led to most of the Internet traffic either using TCP, for applications demanding reliable delivery, or UDP, for applications preferring timeliness to reliability.
This paper focuses on two directions in transport layer research – alternate transport protocols, and multipath approaches – that have materialized to solve the aforementioned problems. Alternate transport protocols, such as Datagram Congestion Control Protocol (DCCP), Stream Control Transmission Protocol(SCTP) and QUIC, were developed as alternatives to the legacy TCP and UDP protocols, aiming to solvesome of their inherent issues in addressing specific application requirements. Multi-path protocols improve single-path protocols’ (e.g., TCP and QUIC) throughput and resilience by leveraging multiple network paths. The 5G feature Access Traffic Steering, Switching and Splitting (ATSSS) specified by 3GPP employs these multi-path transport protocols to utilize both the 3GPP access (e.g., 5G New Radio (NR)) and the non-3GPP access (e.g., Wi-Fi) to provide improved performance. The rest of the paper is organized as follows: in Section 2, we highlight the main issues present in TCP and UDP and describe how the alternate protocols are designed to overcome them. Section 3 provides an overview of the multi-path protocols and discusses their offered improvements. Then, in Section 4, we present results from the testing performed to compare the performance of different protocols in an emulated environment. Finally, Section 5 concludes the paper and summarizes the open research challenges.