DADA: A 2-Dimensional Adaptive Sensor Network Recovery Strategy

It is a common practice to leave a minimum set of sensor nodes active at any particular time while putting others in sleep mode to conserve energy. Although this strategy can substantially extend system lifetime, it makes sensor networks vulnerable to frequent node failures. The failure of an active node can cause service disruptions because it leads to periods of time during which either a portion of the system cannot collect data or the system cannot deliver the collected data to the applications.

On the other hand, sensor nodes are relatively short-lived. Firstly, a node's longevity is limited by the duration of the battery, which is often around months. Secondly, sensor nodes are usually densely built to reduce the sizes/costs, which will cause higher hardware failure rates. Thirdly, sensor applications often require deployment in harsh environments, which can quickly wear out these devices. Fourthly, a node may even appear unavailable to the rest of the network if its communication links are broken, such as the situation in the presence of physical obstacles. Frequent node failures can cause disruptions in both sensing and forwarding.

DADA is targeted at energy-efficient system recovery in the case of failures, by carefully scheduling the sleep/wakeups of the redundant nodes. Leveraging prior experience in CPU scheduling, we designed two algorithms: Rotatory Sentry (R-Sentry) and Persistent Sentry (P-Sentry), wherein sentry nodes can wake up on time to replace the failed active nodes with minimal energy cost:

• P-Sentry. P-Sentry keeps a small subset of redundant nodes awake, and allows the rest to sleep for longer time periods. Those redundant nodes that stay awake play a role analogous to sentries in the real world in the sense that they monitor the health of the active node and, whenever a fault or failure occurs, they jump in to replace the lost node.

• R-Sentry. In R-Sentry, all the redundant nodes go to sleep, but each will periodically wakeup for a short amount of time to probe the health of the active node. Since each redundant node will have its own period of sentry duty, the resulting scheme is very much like a changing of the guards, and for this reason we call this scheme the rotary sentries scheme.

In support of this effort, we have built our own simulator which can model large sensor networks with thousands of or more nodes. We are now proceeding with prototyping the protocols on a testbed of 30 motes.

Publications:

• S. Yu, Y. Zhang, "R-Sentry: Providing Continuous Sensor Services against Random Node Failures," in Proceedings of IEEE International Conference on Dependable System and Network (DSN), 2007. (pdf)
• S. Yu, A. Yang, and Y. Zhang, "DADA: A 2-Dimensional Adaptive Node Schedule to Provide Smooth Sensor Network Services against Random Failures," in Proceedings of the Workshop on Information Fusion and Dissemination in Wireless Sensor Networks, 2005. (pdf)

Home Research