Synopsis Diffusion


This project focuses on new paradigms for robust aggregation in sensor networks.

Previous approaches for computing duplicate-sensitive aggregates in sensor networks (e.g., in TAG) have used a tree overlay topology, in order to conserve energy and to avoid double-counting sensor readings. However, a tree topology is not robust against node and communication failures, which are common in sensor networks. Synopsis Diffusion is a new approach to in-network aggregation that combines energy-efficient multi-path routing schemes with techniques that avoid double-counting. We have developed a formal framework for decoupling message routing from aggregation, enabling energy-efficient but robust routing of intermediate values during aggregation. We show how the insights from the formal framework enable adaptive routing without expending energy on acknowledgement messages. As a result, the new approach achieves significantly more accurate aggregate answers at minimal energy cost.

Recently, we have extended his work by using the tree-based approach where network conditions are good and when accumulated values are small (so that message loss does not have a large impact on the aggregate answer) and the more robust synopsis diffusion approach otherwise. This approach, called tributary-delta, can yield significantly more accurate answers for the same energy cost compared with using either approach alone. The resulting aggregation topology has an analogy to a river flowing to a gulf, where the aggregation initially proceeds along trees (akin to tributaries) and then switches to multi-path when obstacles are encountered (akin to the delta at the mouth of the river). We have developed techniques for dynamically adjusting the balance between tributaries and deltas, as well as novel algorithms for finding frequently occurring values in a sensor network and for other aggregates.

For publications, see the IrisNet Project publications web page.

Researchers


  • Phillip B. Gibbons

    Graduate Student / Interns
  • Suman Nath (Ph.D. student, Carnegie Mellon, now at Microsoft Research)
  • Amit Manjhi (Ph.D. student, Carnegie Mellon)
  • Zachary Anderson (Ph.D. student, U.C. Berkeley)

    Collaborators
  • Srini Seshan (Carnegie Mellon CS)