Measuring packet reordering

Abstract

The Internet architecture provides an unsequenced datagram delivery service. Nevertheless, many higher-layer protocols, such as TCP, assume that packets are usually delivered in sequence, and consequently suffer significant degradation when packets are reordered in flight. While there have been several recent proposals to create protocols that adapt to reordering, evaluating their effectiveness requires understanding the dynamics of the reordering processes prevalent in the Internet. Unfortunately, Internet packet sequencing is a poorly characterized and understudied behavior. This failing can be largely attributed to the lack of accurate and universally applicable methods for measuring packet reordering. In this paper, we describe a new set of active measurement techniques that can reliably estimate one-way end-to-end reordering rates to and from arbitrary TCP-based servers. We validate these tools in a controlled setting and show how they can be used to measure the time-domain distribution of the reordering process along a given path.