Decentralized Operator Screening Scheme for Near-Data Processing Models Based on Precedence Relationships

This paper addresses the issue of data transmission bottlenecks in traditional database management systems when handling large-scale data. It proposes a filtering scheme for operators under the near-data processing model, based on precedence relations. This scheme conducts an in-depth analysis of the precedence relations in the execution sequence of database operators to identify which ones are more suitable for delegation to intelligent storage devices, thereby reducing unnecessary data migration. The research indicates that this strategy can effectively prevent reverse data transmission and significantly reduce the volume of data transmitted on the bus. This results in a substantial improvement in system processing efficiency and a reduction in energy consumption. Experimental verification demonstrates that the filtering scheme based on precedence relations is significantly superior to the traditional filtering method that relies solely on the ratio of input/output data volume. This provides strong support for the development of more efficient database management systems.