Research On Distributed Data Storage Strategies In Wireless Sensor Networks

Wireless Sensor Networks can integrate information collection, processing and transmission together to support researching the physical world and retrieving data from it for researchers. It can play a significant role in environment monitoring, disaster forecasting, etc. It is a research hot spot and an interdiscipline of computer networking, microelectronic technology and ecological environment science, etc. Wireless Sensor networks are data centric network. Despite sink-based real time data collection model, distributed in-network storage for non-realtime data is another research issue.Distributed data storage in sensor networks does matter with network availability and data reliability. Because of the environment complexity, node resource scarcity and network topology dynamics, etc, researching on distributed storage strategies are faced with too many limitations and challenges. Effective and efficient distributed storage strategies do not only satisfy data discovery, but also achieve load balance. This thesis focuses on data storage and discovery for both location-aware and location-free large scale sensor networks, and Hash storage and routing protocol for location-free sparse network. The contributions include four aspects as following.Sensor networks are resource limited networks, so the energy and computation of single node cannot support complicated protocols. In sensor networks, random generated data producers and data consumers compose of peer-to-peer networks. In order to address how to make a producer and a consumer discover each other, this thesis borrows the idea of light reflection theory and proposes a bouncing track storage strategy. In this strategy, the producer can disseminate its data along its own bouncing track; meanwhile the consumer can disseminate its query along its own bouncing track. In theory, every two bouncing track can intersect with each other in ideal case. At the intersecting node, consumers and producers can exchange data. The bouncing track strategy satisfies data retrieval distance-bounded, storage load balance and operation locality. The simulation shows that the performance of bouncing track is better than double rulings and rumor routing.In order to make the consumers and producers discovery each other in the case that the sensor nodes cannot obtain their location, this thesis proposes C-cast storage routing strategy. This strategy does not need location information, rather than it needs to select only two beacons to establish two contour overlay networks in the network. A contour includes all the nodes that have same hop number to one beacon. For a single node, it belongs to two contours in the network. By using contour overlay networks, producers and consumers can disseminate data along contours. In ideal model, it can guarantee that any producer and consumer pair can discover each other 100%. In random model, it is high probability. The C-cast can achieve similar performance and result as the double rulings, which is location-based.Greedy routing is significant for the data transmission and distributed storage in sensor networks. This thesis classifies the greedy routing into strong greedy routing and weak greedy routing firstly. The existing works are mainly on designing weak routing on geographic coordinate, or designing strong routing on greedy embedding graph. However, these two approaches need too much energy for obtaining geographic coordinate or greedy embedding graph. This thesis proposes a light weight tree-based network embedding (TNEG) method. Based on TNEG, a new weak routing strategy is designed, called TGR. TGR can achieve good performance on path stretch factor and load balance. TGR can be used in sparse location-free network. Based on TNEG and TGR, label-based Hash function can be designed for light weight peer-to-peer data storage and discovery strategies.The storage capacity of a single sensor node is limited. This thesis proposes a new method for dividing storage nodes into balancing groups for many data resource nodes in order to maintain as much as possible data in the network. It is assumed that there are k data resource nodes and n storage nodes in the network. The objective of this thesis is dividing the n storage nodes into k group, and each group responds for storage data for one data resource node, while the sum path length of storage node to data resource node is minimal. This thesis proves that this problem can be reduced to maximum weight bipartite matching problem. As the existing algorithms are not applicable for sensor networks, this thesis proposes random greedy algorithm and Voronoi-based algorithm. The simulation shows the performance of Voronoi-based algorithm is better than the random greedy algorithm and its storage path cost is near the optimal one.To sum up, this thesis researches location-based, location-free data storage and discovery strategy and load balance storage problems, while proposes solutions for them. This thesis has the theory and application value to improve the data management, data reliability, etc.