| Data streams arise naturally in a number of monitoring applications in domains such as networking and financial services, but these applications limit the applicability of the standard relational database technology. Many data stream sources are prone to dramatic and bursty, and CPU capability and memory is limited, so that an overloaded system cannot process all of the input data and keep up with the rate of data arrival; as a result, it cannot asure the quality of service. In this case, load shedding become necessary to let the system provide the up-to-date query responses continuously.Query results in Data Stream Management System should meet the requirements of all kinds of QoSs, of which delay is an important one that most users care about. The DSMS framework which this thesis based on takes both CPU and memory into account. One side it can assure that the load is balanced and the CPU capability is effective.This thesis studies the load shedding for aggregation queries over the data streams, which is based on the control-based QoS adaptation framework. In the case of that the CPU capacity is limited and the memory is over loaded, this technique sheds load from memory in aggregation operation, meanwhile,promising a proper QoS. As there is a single aggregation or several aggregation operations, this technique extends the existed framework to an advanced one which can get the subset of the accurate result instead of getting the approximate ones. It keeps the original components such as the cleaner, scheduler and the load shedder, but also extends the load shedder part. Further, adding new components called window-distributor and aggregate-operator to the frame, to promise the accurateness of the result. It combines the new strategy to the existed frame to make sure that the system can be adaptive in dynamic environments when processing the aggregation queries. The experiments show that the system is outperforms other existing ways on resource utilization and deadline miss ratio. |
PDF Full Text Request