Abstract
Latency is one of the most critical performance metrics for a wide range of applications. Therefore, it is important to understand the underlying mechanisms that give rise to the observed latency values and diagnose the ones that are unexpectedly high. In this paper, we study the Internet delay space via robust principal component analysis (RPCA). Using RPCA, we show that the delay space, i.e. the matrix of measured round trip times between end hosts, can be decomposed into two components: the estimated latency between end hosts with respect to the current state of the Internet and the inflation on the paths between the end hosts. Using this decomposition, first we study the well- known low-dimensionality phenomena of the delay space and ask what properties of the end hosts define the dimensions. Second, using the decomposition, we develop a filtering method to detect the paths that experience unexpected latencies and identify routing anomalies. We show that our filter successfully identifies an anomalous route even when its observed latency is not obviously high in magnitude.
Highlights
Latency is one of the most important performance metrics
Note that anomalous routing paths are the ones that have unexpectedly high round trip time (RTT) due to routing misconfigurations or any suboptimal routing decisions including the ones for load balancing
One way to eliminate the measurements with high queuing delays is to aggregate individual client IPs to their Border Gateway Protocol (BGP) prefixes as follows: for each server node we use the minimum RTT measured to any of the client IPs that belong to that prefix
Summary
Latency is one of the most important performance metrics. The quality of a wide range of applications, such as server selection in content delivery networks (CDNs), video streaming, and voice over IP, as well as any time-critical applications, require low latency on Internet paths. We summarize our contributions in this paper as follows: a) we show how to decompose a latency matrix via a recent technique, RPCA, into a low-rank and an inflation component; b) we investigate the features of end hosts that result in the low-rank property and we find that both geolocation and the AS of the end hosts define the dimensions of the delay space; c) we propose a method to diagnose the inflated paths by the inflation-toestimated-latency ratio filters; d) we show that our filter successfully pinpoints the routing anomalies even when the RTTs on the paths are not obviously large; e) we show that our filter successfully pinpoints the routing anomalies even when all measured paths between the end hosts from two regions are inflated; and f) we show how to apply our filter in the case that RTT measurements between some end hosts are missing.
Sign-in/Register to view highlights & summary 
Published Version (
Free)
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call