An Empirical Characterization of Internet Round-Trip Times

Abstract

End-to-end round-trip times (RTTs), which measure the time when the source transmitted data and when it received confirmation that the data was received, have been used by several Internet applications and protocols as a way to estimate network load and congestion. However, the Internet's ever increasing size and complexity pose many challenges to the study of the RTT. RTT's stochastic nature combined with diverse network topologies, technologies, and workloads are part of the problem, as well as difficulty in acquiring representative RTT samples or testing how RTT measurements are affected by changes in protocols. As part of the answer to these challenges, this paper presents a characterization study of RTT traces collected from both real- as well as simulated networked environments. We verify that temporal- and spatial factors cause RTT behavior to exhibit particular trends. Using rigorous analytical methodology, we also confirm that RTT distributions can be modeled as a power law. We then use RTT power law statistics to validate and fine-tune simulation environments.