Reducing Latency in Internet Access Links with Mechanisms in Endpoints and within the Network

Excessive and unpredictable end-to-end latency is a major problem for today’s Internet performance, affecting a range of applications from real-time multimedia to web traffic. This is mainly attributed to the interaction between the TCP congestion control mechanism and the unmanaged large buffers deployed across the Internet. This dissertation investigates transport and link layer solutions to solve the Internet’s latency problem on the access links. These solutions operate on the sender side, within the network or use signaling between the sender and the network based on Explicit Congestion Notification (ECN). By changing the sender’s reaction to ECN, a method proposed in this dissertation reduces latency without harming link utilization. Real-life experiments and simulations show that this goal is achieved while maintaining backward compatibility and being gradually deployable on the Internet. This mechanism’s fairness to legacy traffic is further improved by a novel use of ECN within the network.