LUNCHBUNCH! Grad Student Presentations

Dependable Networking in Spite of Dynamics and Scale

Hongwei Zhang
OSU Graduate Student

Thurs., June 2nd
11:30am; 480 Dreese Labs
All interested parties are invited.
Pizza lunch will be served.

Abstract:
Computer networks evolve in both the underlying communication techniques and the scales. Along with opportunities, the evolution brings with it new challenges to network dependability. In this talk, I will focus on two aspects of the issue: First, I present how the properties of wireless communication affect the design of routing and transport protocols in wireless sensor networks; Second, I discuss how the large network scale affects the design of fault-tolerant protocols.

On routing in wireless sensor networks, I show the impact of such factors as environment, packet type, packet length, and interference pattern on wireless link quality, and I propose a beacon-free routing protocol Learn-on-the-Fly (LOF) that estimates link quality solely based on data traffic. For reliable and real-time transport in sensor networks, I discuss the impact of application traffic pattern on error as well as contention control, and I propose a transport protocol Reliable-Bursty-Convergecast (RBC) that innovates the techniques of window-less block acknowledgment and priority-based contention control. Both LOF and RBC have been deployed and verified in the field sensor network experiment ExScal where around 210 Stargates and 1,000 motes were deployed.

On the design of fault-tolerant protocols in large scale networks, I focus on the problems of locally containing the impact of faults and quickly stabilizing networks from fault perturbations in a stable manner. On fault containment, I discuss - in the context of clustering and routing - techniques that take advantage of geographic information, spatial and temporal redundancies in networks. For stable and quick stabilization, I discuss - in the context of path-vector routing - techniques that enhance fault-awareness and resolve uncertainty in fault detection.