Research On Database Acceleration On API And Related Storage Engines

With the rapid development of Internet technology,the performance requirements for efficient retrieval of large amounts of information are increasing.New requirements have been put forward to respond to various queries quickly,so it is particularly necessary to explore a universal and reliable database acceleration API that supports software and hardware co-processing for database systems.Based on a heterogeneous system composed of CPU and FPGA,this thesis designs a layered database acceleration API,and a storage engine that is compatible with MySQL database and also supports hardware acceleration.The research work is mainly divided into the following three aspects.(1)Analyze the basic ideas and main ways of database hardware acceleration,and combines the characteristics of mainstream relational database MySQL,this thesis propose a software architecture that pushes some SQL statements down to the FPGA to perform calculations to achieve database query acceleration.(2)Design and implement a storage engine that supports SQL offloading for the query statements of the database MySQL.The storage engine supports selection-based filtering and projection operation processing,and can parse SQL statements into truth tables and other forms for hardware identification(3)The database acceleration API is designed and implemented to provide users with a set of acceleration APIs.It also implements unified management,status monitoring,and task scheduling of hardware devices,and builds a database operation path from CPU processing to FPGA calculation.The test platform is completed under the Linux system.The results show that the API can complete the filtering and projection operation,and dynamically use FPGA resources.The storage engine can accurately parse the offloading SQL and complete basic table operations.Combined with the API,the storage engine can ensure that the client gets the correct query results and meets the expected requirements.