Hardware Acceleration For Relational Databases On FPGA

Database is an important tool for data management.In the era of ”big data”,mobile communication,e-commerce,telemedicine,health management,and other applications will generate a huge amount of online data.Traditional databases cannot satisfy these application's requirements of large data volume,diversified data structure on a short processing time.With the development of semiconductor technology,the requirements of the database for computing ability cannot be limited to the isomorphism computing architecture based on Von Neumann architecture.To break off the barriers between the processing ability of homogeneous computing patterns and the growth rate of data,the heterogeneous parallel computing architecture with a new generation of high-performance computing chips attracts people's attention,to achieve higher performance.Hardware acceleration on Field Programmable Gate Array(FPGA)is used to accelerate database operations,and FPGAbased heterogeneous acceleration system architecture of relational database has become a very valuable research direction.Data analysis is manipulated by structured query language on relational databases.Sort operation and join operation are the two most important operations on databases.However,these operations are time-consuming and data-intensive,the offloading of these operations on FPGA has been a technique of growing interest for a long time.This thesis proposed a method of sorting operation and join operation based on FPGA,furthermore proposed a heterogeneous accelerating architecture for different relational databases.The main contributions of this paper are as follows:A modular sorting matrix for the sort-based accelerator.Sorting is one of the most fundamental problems in mathematics and computer science.The sorting units used in databases must be reasonably simple to maintain while having high speed and being flexible for a wide range of operating conditions.Hence we present efficient techniques for designing a high-throughput sorting matrix for continuous variable-length data sequences.This matrix consists of identical basic sorting cells.The sorting cells work in pipeline and parallel modes,and the matrix can simultaneously process multiple data streams,which can be combined into the high-width single-channel data stream or low width multiple-channel data streams.What's more,an equi-join and filter operation based on this sorting matrix are also implemented on FPGA.With the proposed sorting matrix,the accelerator can realize sorting,sort-merge joining,and filter operations.Two Hardware accelerator for hash join algorithms.The hash join algorithm is one of the most popular join algorithms on relational databases,and the non-uniform distribution of data caused by hash collisions and the methods of hash table accesses negatively affects the performance of the hash join.Based on these optimization directions,two hashing join architectures are proposed.For the scenario that the join attribute is not unique,two hash functions are employed to distribute data as evenly as possible,and then multiple entries of the collision list are provided;finally,a small number of the static random-access memory(SRAM)of the FPGA are used to conduct the join operation in parallel.For the scenario that the join attribute is unique,a non-collision parallel hash join architecture is proposed,this architecture utilizes multiple hash functions and content-addressable memories(CAMs)to eliminate hash collision,thereby ensuring a worst constant memory access for each phase in the hash join and consequently improving the hash join throughput.A Hybrid database hardware accelerate architecture for different relational databases.Different heterogeneous technologies have their own advantages in different scenarios.When a heterogeneous platform is used to accelerate the database,the delay of data analysis results sending should be minimized and the data of processing should be improved.Scalability and flexibility are also important in system design.To resolve this problem,this paper proposed a general database acceleration system architecture based on FPGA.For the hardware,this architecture puts forward targeted optimization methods for different data operations.For the software,it provides a unified call interface for different database software,which makes it loosely-coupled with database software and more widely applicable.Through multiple experiments on the standard benchmarks,the correctness and effectiveness of the proposed architecture are proved.