| Summary: | Multimedia applications are known to have low end-to-end latency requirements for interactivity, suggesting UDP to be a suitable transport protocol. Traditionally, the bandwidth requirements were set to be well below the network capacity. Modern applications could evolve if more bandwidth were available without giving up interactivity. However, the lack of a suitable transport providing such high bandwidth with low latency has prevented the existence of the next generation of multimedia applications. A transport built on top of UDP can no longer traverse the network without being hit by major packet drops or severely damaging other traffic. TCP, having built-in congestion control, is safe to use for high bandwidth requirements, but has been associated with unexpected high latencies.
In this thesis, we go against conventional wisdom and design Paceline, a transport layer on top of TCP to support high-bandwidth low-latency applications. Paceline improves latency through application level rate control. It also detects when TCP connections exponentially back-off and automatically replaces them with standby connections. Through evaluation we show Paceline improves TCP's shortcomings in terms of end-to-end latency in extremely congested networks, without sacrificing TCP's effectiveness in utilization or bandwidth sharing.
We design two different applications on top of Paceline for streaming live video. We present a preliminary evaluation of the application's performance when using Paceline in congested networks and show significant improvements compared to TCP. === Science, Faculty of === Computer Science, Department of === Graduate
|
|---|
