Analysis And Improvement Of TinyDB: A Prototype Of Wireless Sensor Network Database

Large-scale wireless sensor networks are integration of sensor techniques, embedded computation techniques, distributed computation techniques and wireless communication techniques, which have a wide application future. It is a new area to architect sensor networks as virtual databases with energy-efficient data management and acquisition.According to TinyDB, the most famous wireless sensor network database prototype, this paper first analyzes its full source codes for laying groundwork for future system migration and improvement. TinyDB system consists of two parts, one front-end and lots of embedded database engines. The front-end of this kind database is an interface between applications and sensor networks, which represents wireless sensor networks as unique relation databases. Applications retrieve data from the only table named sensors that map each kind of sensor networks data to one field. The front-end builds messages to load context of many kinds of queries, e.g. event-based queries, and then disseminates them sequentially into the sensor networks. The front-end is also responsible for handling all coming query results and dispatches them.The other part of this database needs an embedded database engine running inside every sensor network node. This thesis studies issues on design and implementation of the engine, including data schema, distributive execution of queries, in-networking aggregation and grouping, and so on. During each step of query process, nodes collaborate to produce results in every epoch, and forward them up to database front-end. There is also some optimizing done by the engine to improve performance.In the third part of the thesis a distributive join algorithm is described, which can support complex data filtering and event-detection in many surveillance applications. It assembles complex predicates into one static table that is pushed into and stored inside nodes. This algorithm performs join queries over sensors table and static tables in-networking. Both algorithm analysis and experiments show data traffic reduce obviously if the selectivity of static table is low.At last, the primary work is summarized, and some future works are stated.