Balanceamento de carga em nível de aplicação para redes de sensores sem fio reativas: uma abordagem baseada em problemas de otimização com restrições

Abstract

In wireless sensor networks (WSNs), the overload of nodes can result in the degradation of network performance, requiring the adoption of dynamic load balancing techniques to prolong the service life and availability of its services. Currently, these networks are increasingly being used in applications with high demand for video and image processing, which means that the processing load, as well as the communication load, must be considered in load balancing. Recently, techniques based on heuristics have been proposed to balance the load at the application level. However, these do not have any quality guarantee and since they are distributed, they require communication, which can affect the network’s lifetime. Thus, this work presents a new solution that is based on techniques of Constraint Optimization Problems to solve the load balancing in WSNs at the application level.The proposed approach explores the limited redundancy in the coverage areas and employs a semi-distributed solution, avoiding the complexity of distributed algorithms. The proposed approach ensures that the detected events are processed during the network’s lifetime and that a fixed number of messages will be exchanged for each new detected event. In order to minimize the negative impact on the lifetime, the COP-Feedback approach incorporates propagation control procedures that prevent the sending of redundant messages during the load balancing process. Experiments point out that the proposed approach allows the network to perform better than the heuristic ones considered state-of-the-art, both with nodes arranged in a grid, as well as with nodes randomly arranged. The results also showed that COP-Feedback maintains its behavior both in high-density as well as low-density networks, which is not the case with heuristic techniques. In addition, COP-Feedback presented results close to that of a greedy solution, used as upper-bound in the analyzes, but which cannot be used in real scenarios because it requires centralization of the state of the network nodes.