Dvfs-based Energy-efficient Query Processing In DBMS

In recent years, the energy-efficiency has become an emerging research topic in power-aware DBMS. Dynamic voltage and frequency scaling (DVFS) is an effective method for saving energy by dynamically scaling CPU frequency. This paper tries to explore the potential relationship between performance, energy consumption and power of query operations in DBMS under different ACPI regulators, to find out energy-efficient characteristics of operations, and design the improved grace hash join (GHJ) and block nested-loop join (BNL) algorithms by leveraging Userspace regulator, which aims to check the feasibility of energy-efficient join algorithms based on DVFS. The contributions of the thesis are as follows:(1) We profiled the power characteristics of22queries from TPC-H benchmark under different ACPI governors for PostgreSQL DBMS, and grouped them into different categories according to their different energy and performance characteristics, and analyzed their energy efficiency by looking into different query types and their operators in their respective query plans.(2) We analyzed the power characteristics of two typical join algorithms in traditional RDBMS, i.e., the block nested-loop join and grace hash join algorithms, and proposed their corresponding energy-efficient variants that is regulated with DVFS methods in Userspace. For BNL, we improved its energy-efficiency by overlapping its I/O operation with its CPU operations; while for GHJ, we re-design the algorithm by prolonging the contiguous I/O interval or CPU operation interval, which validates the DVFS regulation method and improves the energy-efficiency.The experimental results show Userspace regulator-based GHJ and BNL achieved good energy-efficiency, it can be concluded that energy-efficiency of query operations can be achieved by DVFS.