On the Constancy of Latency at the Internet's Edge

Latency is a key determinant of application performance and, most importantly, end-user experience. While a low latency value between any pair of endpoints will immensely benefit the applications or data transfers between those end-points, whether that end-to-end latency fluctuates and, if yes, to what extent also have crucial performance implications. Recently, Davisson et al. replicated a decade-old seminal work on assessing the constancy of latency between arbitrary endpoints in the Internet. In this work, we follow the approach of Davisson, albeit with a focus on end users: We assess the constancy of latency between end users and the servers that serve most of the content the end users typically consume. These are the endpoints that are of interest. We measure latencies between RIPE Atlas anchors and the edge servers of various content delivery networks (CDNs), both selected from diverse geographic locations and networks. We show that end-to-end latencies between end users and CDN edge servers are stable, with few significant spikes or variations. We find that the durations without change for the analyzed paths are much larger, with median maximum values at least three times greater than those in previous studies on RIPE Atlas anchor paths. We release our tools and analyses to foster open and reproducible research.