LIFEGUARD: Locating Internet Failure Events and Generating Usable Alternate Routes Dynamically

The Internet is far from achieving the “five 9s” (99.999%) of availability provided by the public telephone network. Based on anecdotal evidence as well as a first-of-its-kind measurement study assessing routing failures affecting Amazon’s EC2 data centers, we find that network outages are surprisingly common and highly problematic. These outages incur large costs, disrupt critical services and applications, and are generally difficult to detect and isolate. In this thesis we argue that it is possible to isolate wide-area network faults at a routerlevel granularity without requiring control over both end-points. We present LIFEGUARD, our prototype system for Locating Internet Failure Events and Generating Usable Alternate Routes Dynamically. LIFEGUARD leverages distributed network measurements, a continually updated historical path atlas, and novel measurement tools such as spoofed forward traceroute to infer the direction of failures, isolate faults, suggest alternate working paths, and provide views into routing behavior from before the onset of outages. We evaluate LIFEGUARD with a four month study monitoring real outages on the Internet, and present specific examples and aggregate characteristics of the failures we observed. Presentation of work given on June 2nd, 2011 Thesis and presentation approved by: