Enhancing speed of SQL database operations using GPU

Graphic Processing Unit (GPU), has proved to be an efficient co-processor in the field of conventional computations. Dramatic acceleration has been achieved in the earlier work on different database operations on GPUs which were not part of conventional database languages like SQL. While GPUs were designed for geometric primitive's visualization, it can also be used to execute database operations efficiently by using inherent pipelining and parallelism, multi-threaded architecture, vector processing functionality of GPUs along with Single Instruction and Multiple Data (SIMD) capabilities to evaluate semi-linear queries based on attributes. The efforts required by database administrators to learn languages like CUDA or change modules in order to provide aid to libraries which are not present in SQL is reduced. This research work focuses mainly on developing a system to enhance the execution speed of SELECT queries and how effective it would it be in comparison to traditional methods. This work intends to provide a clear portrayal of how GPU hardware can be used for query execution in the area of relational databases in future. The algorithms were implemented on databases consisting of nearly a million records with the aid of a programmable GPU. The existing results suggests that using GPU as a co-processor can significantly improve execution of database operations over optimized CPU-based implementation.