On the latency and routing impacts of remote peering to the Internet

Abstract

Remote peering (RP) has crucially altered the Internet topology and its economics. In creasingly popular thanks to its lower costs and simplicity, RP has shifted the member base of Internet eXchange Points (IXPs) from strictly local to include ASes located any where in the world. While the popularity of RP is well understood, its implications on Internet routing and performance are not. In this thesis, we perform a comprehensive measurement study of RP in the wild, based on a representative set of IXPs (including some of the largest ones in the world, covering the five continents). We first identify the challenges of inferring remote peering and the limitations of the existing methodologies. Next, we perform active measurements to identify the deployment of remote IXP inter faces and announced prefixes in these IXPs, including a longitudinal analysis to observe RP growth over one and a half years. We use the RP inferences on IXPs to investigate whether RP routes announced at IXPs tend to be preferred over local ones and what are their latency and latency variability impacts when using different interconnection meth ods (remote peering, local peering, and transit) to deliver traffic. Next, we asses the RP latency impact when using a remote connection to international IXPs and reaching prefix destinations announced by their members. We perform measurements leveraging the in frastructure of a large Latin American RP reseller and compare the latency to reach IXP prefixes via RP and four Transit providers. Finally, we glimpse some of the RP impli cations on Internet routing. We evaluate how RP can considerably affect IXP members’ connection stability, potentially introduce routing detours caused by prefix announcement mispractices and be the target of traffic engineering by ASes using BGP communities.