Query Processing in Sensor Networks with Dynamic Execution Plan Refinement

A (wireless) sensor network can be initialized with one big cluster. This cluster contains all nodes. In addition, it has a head node. The head is nearest to the nearest point of all cluster members. Members communicate with their head using respective local path. Local paths can be multi-hop. A head caches all data from its members. The cache life is adjusted based on how accurate it was previously. This introduces the concept of local data dynamicity. If cluster data becomes too dynamic, it is decomposed into multiple smaller clusters. This reduces cache update cost and localizes dynamic zones. Other low dynamic regions still benefit from longer cache life. When there are multiple clusters, a cluster head uses a multi-hop global path through other cluster heads to communicate with the base station. At the base station, after a query is parsed and syntactically optimized, it is classified. The static query optimizer then uses this classification to choose the initial query execution plan. The plan combines both query and routing strategies. The static query optimizer also sets the plan refinement guidelines. Cluster heads use these guidelines to dynamically refine the current execution plan while queried by the base station. A base station can only query a cluster head. The cluster head can in turn provide information about its cluster members.