Research On Adaptive Query Optimization And Scheduling Strategy On Data Stream

Recently more and more research communities focus on a new class of data-intensive applications in which the data is modeled best not as persistent relations but rather as transient data streams. Examples of such applications include financial applications, network monitoring, security, telecommunications data management, web applications, manufacturing, sensor networks, and so on. In all of the applications cited above, it emphasizes that data arrival in multiple, rapid, time-varying and requires decisions are acted in close to real time. So it must be considered the memory and the response time of the system. Traditional database technologies cannot satisfy with it. So the research on the data stream technologies have been attracted more and more attentions in the database field.Query on the data stream usually is Continuous Query for the data comes continuously with time. During the query processing, many factors are changing, for example memory usage, selectivity of operators, the rate of the data stream and so on. To improve query performance and in response to changing conditions, adaptivity becomes critical to a data stream systems as compared to a traditional DBMS and has attracted more and more attentions.In this dissertation, we discuss query processing mechanisms in DSMS and propose a adaptive query optimum scheduling approach, namely MultiFactor, which can changing the query plan while the characters of data stream (e.g. rate, selectivity and execution time) and performance of query (e.g. memory usage and output delay) are altering. It decides a query plan by operator's processing speed that is the number of tuples it consumed per unit time. Selectivity and processing speed of operators each are used to decrease memory usage and reduce response time. And the speed of stream, which is important for the system performance, is also considered in this strategy. We also research into the scheduling strategies in the data stream and propose one that executes the query plan by unequal time slice. The time-slices are determined by deadline time which can be declared in a query sentence.Characters of data stream (e.g. selectivities, the numbers of tuples processed per unit time and speed of stream) could be learned during query execution by gathering statistics over a period of time. We consider that the query executing time is composed of a set of restrictive time windows. The system collects statistics independently in each window and computes them. Then the system use the results from the ith window to decide whether the query plan for the (i+1)st window will be changed.The parameters of data stream (e.g. rate of data stream, selectivity and processing speed of operator) are tested at last. This dissertation also gives the algorithms of MultiFactor, describes the result of experiments and compares the performance of the various operator-scheduling approaches described in this paper. Experiment has proved that this approach reduces output time latency and memory requirements. And it is better than other algorithms.