CAREER: Exploiting Tomography in Network-Aware Protocols: Theory and Practice

Purdue University, West Lafayette IN

Investigators

Abstract

The overarching goal of our research is to address the problem of scalable and network-adaptive overlay networking, including load balancing, optimized dispersity routing, and low-overhead network monitoring. To achieve this goal, we will design novel techniques for spatially and temporally exploiting end-to-end measurements for inference and on-line adaptation. Our techniques will be implemented and evaluated in a system that includes three components: 1. Overlay networks for multicast, anycast, geocast, and peer-to-peer services: We will design methods to infer underlying network characteristics (e.g., topology, losses) and host properties (e.g., location, remaining battery power), and exploit them for adapting overlay trees and meshes on the Internet. We will also model and characterize the structure, dynamics, and performance of various overlay network classes. Understanding and isolating the impacts of Internet topology and the overlay construction algorithm will be an integral component of this work. 2. Load balancing, optimized dispersity routing, and coordinated congestion control: We will study the feasibility of using correlations among delay/loss measurements to partition TCP or application-layer flows according to inferred bottlenecks in real time. The resultant flow partitions can be used to continuously balance loads, utilize dispersity routing and parallel paths, and make coordinated congestion avoidance decisions. 3. Distributed network monitoring: We will design lightweight tomography-based methods to continuously diagnose network problems and detect anomalies or attacks based on inferred network characteristics. Based on our solutions and the associated studies of (1) network-layer and application-layer interactions; (2) overlay properties, cost, and dynamics; and (3) the feasibility of on-line tomography, we will be poised to expand our scope from overlay networks, to other challenging problems in network security, and wireless and sensor networks. The proposed research covers various facets of networking (from theory to implementation) and will provide graduate students with a balanced exposure to both theory and practice that will be invaluable in their future careers. In addition, the research will be integrated with lab and seminar development and mentoring and outreach activities that span levels from high school to graduate students. The programs aim at increasing *women* student participation and retention statistics, by attracting students to quality networking research. A set of web-based graphical user interfaces and lab assignments for network simulation and measurement will be developed, leveraging both our proposed research software, and tools available from research labs and universities. These labs will aid high school and undergraduate students appreciate the scale, heterogeneity and dynamics of the Internet in an interesting environment. Undergraduate and graduate mentoring, and research seminars with industry involvement, will encourage students, especially women, to pursue research to the PhD level. All labs and seminar material will be made freely available on the Web for the benefit of similar programs at other institutions.