Performance analysis of pluggable congestion control in QUIC protocol

Abstract

The vast majority of internet infrastructure operates using HTTP/2.0 with Transport Layer Security (TLS) on top of the transport layer that uses Transmission Control Protocol (TCP) for data transfer. One of the most critical problems with this setup is that upgrading the transport protocol requires us to update all middle boxes across the internet infrastructure. This means that there is minimal scope for TCP to evolve. There are also certain other areas where the existing stack makes compromises in performance such as performing slow handshakes, forcing sequential packet numbers globally and failing to account for the nature of wireless networks. To overcome these significant shortcomings, Google developed a protocol named Quick UDP Internet Connections (QUIC). It is an application layer protocol that uses UDP for its transport. By moving the ‘reliability’ aspects of TCP from the transport layer to the application layer, QUIC enables faster developments and even allows multiple variations of the protocol. In this paper, two congestion control algorithms for QUIC are proposed with different architectures - an event-based model that handles wireless traffic better than the existing system, and a performance-based model that controls congestion by measuring absolute network performance. The performance of QUIC with each congestion controller has been analysed through tests conducted in controlled network simulations, under varying bandwidth and induced packet losses, to mirror the conditions faced over the internet. It is observed that the proposed congestion control algorithms outperform the current solution in all test scenarios.