Abstract

A range of performance metrics beyond throughput are increasingly becoming relevant for user experience. Notably, latency under load—the end-to-end latency of an Internet path when the network is loaded with traffic for a period of time—is a distinguishing feature, as increased latency, even for a short period of time, can disrupt connectivity for a wide range of applications. In this brief position paper, we use preliminary experiments from a home broadband measurement testbed across Chicago to demonstrate that latency under load can differ significantly, both for users across different speed tiers and for users within the same speed tier. We use this position paper to present a few compelling examples, to seed a discussion about ways to measure and compare latency under load across subscribers.