Enhancing end-to-end availability and performance via topology-aware overlay networks

  • Authors:
  • Junghee Han;David Watson;Farnam Jahanian

  • Affiliations:
  • Samsung Electronics Co., Ltd, Dong-Su-Won, 416 Maetan-3dong, Yeongtong-gu, Suwon-si, Gyeonggi-do, Korea;Ford Motor Company Dearborn, MI 48121, United States;University of Michigan, Department of EECS, 1301 Beal Avenue, Ann Arbor, MI 48109-2122, United States

  • Venue:
  • Computer Networks: The International Journal of Computer and Telecommunications Networking
  • Year:
  • 2008

Quantified Score

Hi-index 0.00

Visualization

Abstract

This paper proposes a novel overlay architecture to improve availability and performance of end-to-end communication over the Internet. Connectivity and network availability are becoming business-critical resources as the Internet is increasingly utilized as a business necessity. For example, traditional voice and military systems are turning into IP-based network applications. With these applications, even short-lived failures of the Internet infrastructure can generate significant losses. To satisfy these needs, the concept of overlay networks has been widely discussed. However, in the previous studies of overlay networks, a measurable number of path outages were still unavoidable even with use of such overlay networks. We believe that an overlay network's ability to quickly recover from path outages and congestion is limited unless we ensure path independence at the IP layer. Hence, we develop a simple but effective overlay architecture increasing path independence without degrading performance. The proposed overlay architecture enhances prior studies in the following ways: (1) we deploy overlay nodes considering topology and latency information inside an ISP and also across ISP boundaries; (2) we use a source-based single-hop overlay routing combined with the above topology-aware node deployment; (3) we increase the usage of multi-homing environment at endhosts. In this framework, we develop measurement-based heuristics using extensive data collection from 232 points in 10 ISPs, and 100 PlanetLab nodes. We also validate the proposed framework using real Internet outages to show that our architecture is able to provide a significant amount of resilience to real-world failures.