On the Effectiveness of Measurement Reuse for Performance-Based Detouring David Choffnes and Fabi´ an E. Bustamante Department of Electrical Engineering & Computer Science, Northwestern University {drchoffnes,fabianb}@cs.northwestern.edu Abstract— For both technological and economic reasons, the default path between two end systems in the wide-area Internet can be suboptimal. This observation has motivated a number of systems that attempt to improve reliability and performance by routing over one or more hops in an overlay. Most of the proposed solutions, however, fall at an extreme in the cost-performance trade-off. While some provide near-optimal performance with an unscalable measurement overhead, others avoid measurement when selecting routes around network failures but make no attempt to optimize performance. This paper presents an experimental evaluation of an al- ternative approach to scalable, performance detouring based on the strategic reuse of measurements from other large dis- tributed systems, namely content distribution networks (CDNs). By relying on CDN redirections as hints on network conditions, higher performance paths are readily found with little overhead and no active network measurement. We report results from a study of more than 13,700 paths between 170 widely-distributed hosts over a three-week period, showing the advantages of this approach compared to alternative solutions. We demonstrate the practicality of our approach by implementing an FTP suite that uses our publicly available SideStep library to take advantage of these alternative Internet routes. I. I NTRODUCTION Path selection in the wide-area Internet is known to be suboptimal in terms of end-to-end latency, loss rate and TCP throughput. Building on a large body of previous work measuring the behavior of Internet routing, the Detour study showed the potential benefits of detouring flows via a third node [31]. Since then, there has been a number of proposed overlay routing systems that attempt to improve reliability and performance [5], [14], [30], [36]. Most solutions fall at either extreme in the cost-performance trade-off: RON provides near- optimal performance at the cost of measurement overhead that is quadratic in the number of nodes in the system, whereas Gummadi et al.’s technique requires no measurement overhead to route around network failures, but does not attempt to optimize performance. In this paper, we present an experimental evaluation of an alternative approach to scalable, performance detouring. The proposed approach relies on the strategic reuse of mea- surements from other large-scale distributed systems, namely content distribution networks (CDNs). CDNs cache copies of web objects on thousands of servers worldwide and redirect clients to different servers, over short time scales, based on server load and network conditions [2]. In previous work [35], we demonstrated that these redirections are primarily driven by network conditions and suggested that they could be used to identify quality Internet paths without additional monitoring. This paper presents a thorough evaluation and comparison of SideStep, a detouring service we have implemented based on these ideas. SideStep employs CDN redirections as hints on network conditions, to efficiently identify higher performance paths with little overhead and no active network measurement. Paraphrasing Lampson [18], a hint is the saved result of some measurement or computation used for the purpose of making a system run more efficiently. Since hints may be wrong, there must be a way to check their correctness before taking any unrecoverable action. SideStep employs an effective, low-cost strategy for validating the quality of recommended paths. Our work makes the following contributions: • A detailed description of the design and implementation of the SideStep detouring service. SideStep is the first open-access, scalable solution to finding high-quality overlay paths. • Results from a wide-area evaluation of the deployed system, proving that CDN redirection dynamics can be seen as hints regarding high-quality candidate detour points, and that these hints can effectively support a highly scalable detouring service. • An open-source API and library implementing our SideStep detouring service, along with an FTP suite (DraFTP) that relies on SideStep to seamlessly take advantage of alternative Internet routes and serves as a model for other client applications. Before describing our experimental approach and presenting our evaluation results, we briefly describe the design and implementation of SideStep (Sec. II). We then experimentally show the benefit of the SideStep detouring service in terms of end-to-end performance improvements (Sec. IV). Our results show how reusing CDN measurements allow us to eliminate the scalability constraint imposed by actively measuring all overlay paths (e.g., as done in the RON approach). Finally, we demonstrate the practicality of our approach by implementing DraFTP – an FTP suite that uses our portable, publicly available SideStep library to seamlessly take advantage of alternative Internet routes. We discuss the limitations of our approach and challenges for future work in Sec. V and conclude in Sec. VII.