سیاست های مدیریت حسگر برای ارائه نرم افزار QoS

Wireless sensor networks are uniquely characterized by tight energy and bandwidth constraints. These networks should be designed to provide enough data to their application so that a reliable description of the environment can be derived, while operating as energy-efficiently as possible and at the same time meeting bandwidth constraints. These goals are typically contradicting and must be balanced at the point where the application is best satisfied. In this paper, we address the problem of maximizing lifetime for a wireless sensor network while meeting a minimum level of application quality of service. This maximization is achieved by jointly scheduling active sensor sets and finding paths for data routing. significantly increased through optimized several heuristic methods. Simulation results show that several heuristic policies can achieve near optimal network lifetime.

Interest in the use of wireless sensor networks has blossomed over the last several years due to technological advances enabling smaller devices and the realization of the potential benefit of such networks in many applications. In some situations, sensor networks may consist of sensors with overlapping coverage areas that provide redundant information, giving an application a quality level that is more than necessary. Rather than provide this unnecessary redundant data, it may be desirable to reduce power consumption and conserve energy in these sensors to lengthen the lifetime of the network or minimize the rate at which the sensors must be replenished with energy. This energy conservation can be accomplished through a number of methods. For example, sensors’ reporting rate or data resolution can be adjusted, or the sensors can be turned off completely for an extended period of time. Balancing the application quality with this goal of energy-efficiency essentially provides a type of application quality of service (QoS). To efficiently provide this QoS to the application, interaction with lower levels of the sensor network’s protocol stack is required. Recently, efforts have been made to develop middleware providing this interaction while simplifying software development efforts [1] and [2]. Here, we discuss the advantages of efficient sensor management when used in such a middleware system. In this work, we show how the use of two strategies––turning off redundant sensors and using energy-efficient routing––can be used to extend network lifetime while meeting a required level of application quality. Recent research has focused on methods of in-network data aggregation to reduce the amount of communication in dense wireless sensor networks. In this case, low-level fusion is typically performed on data from neighboring sensors before being sent to a data sink. As an alternative to this approach, redundant sensors can be turned completely off for periods of time to save energy. Of course, there is a tradeoff between power consumption and overall quality of the data delivered by the network when choosing which approach to use. We consider the latter approach in this work but realize the benefits of the former. In our work, we also take careful consideration at the routing layer, calculating routes in conjunction with the sensor scheduling. We show in this paper how to jointly optimize sensor scheduling and data routing to extend the lifetime of sensor networks and we analyze the performance of some simple heuristic sensor management and routing policies. The rest of this paper is organized as follows. Section 2 gives the problem formalization. Section 3 describes the simple policies that we use for scheduling and routing in our simulations. Section 4 provides simulation results. Section 5 provides context for where intelligent scheduling/routing can be used. Section 6 addresses related work. Section 7 concludes the paper.

Intelligent sensor management is one way to provide quality of service to an application. We have formalized a sensor management problem and shown through simulations that for applications using spatially diverse sensors, several heuristic scheduling/routing approach can achieve near optimal lifetime. In our model, average power consumption of a node to route data is proportional to the routing load. In real networks, this power cost may actually depend on the state of the node. For example, in an IEEE 802.11 network, the marginal power cost to route data for a node already being used as an active sensor in the current set is significantly less than for one that is otherwise idle. For the active sensor, the energy cost is approximately proportional to the difference between idle power and the average of the receive power and transmit power. Meanwhile, a sensor that is not being used in the current sensor set would otherwise have its radio turned off, and the marginal cost is proportional to the sum of the receive power and transmit power. Similarly, if a node is being used to route data in multiple paths, its energy cost per forwarded packet is significantly less. When modeling such types of networks, the problem can no longer be modeled as a generalized maximum flow problem and becomes NP-hard, requiring a heuristical approach to solve. While not representative of all networks, we believe that our model does represent some typical networks that are likely to be used in sensor networks. Many TDMA-based networks can be modeled as having power dissipation that is nearly proportional with the routing load, as we have modeled. Such networks include Bluetooth networks when an intelligent scheduling method is used. Still, even in these TDMA-based networks, traffic schedules may need to be set up to allow for this efficiency, and this overhead is not accounted for in our model. In all simulations, we did not consider the overhead of setting up and tearing down routes. We acknowledge that in real situations this could in fact impact average power consumption and network lifetime and our model would need to account for these additional factors before running the optimization program. It should also be noted that these simulations were simplified and that quirks of specific routing protocols and other factors could cause discrepancies in performance measurements. Our approach also requires global information about the neighborhoods of each node. For small networks, this may not be a problem, but requiring this information to be propagated back to the base station would not scale well for larger networks. The motivation for this work was a larger project in which we are developing a middleware system for use in wireless sensor networks. We will incorporate the ideas and strategies presented in this paper during the development of this middleware. We are also developing distributed algorithms in which sensors use information from neighboring nodes, gathered through eavesdropping, to determine what their current network role should be. We hope that these algorithms will achieve lifetime results that are near the upper bounds that can be found by the methods described in this paper.