Parallel Query on the In-Memory Database in a CUDA Platform

With the progress of the manufacturing processes for GPUs, the number of cores in a GPU was significantly increased. This not only brought people a better visual effect in graphic applications, but also made people enable parallel computing gradually using GPUs instead of CPUs. The CUDA (i.e., a GPU-based parallel computing architecture) developed by NVIDIA has been used in various fields for several years. In this paper, a database query system on a CUDA platform is realized. We implement all relational algebra operators as well as data sorting and data grouping provided by the SQL specification. In the experiments, we not only observe both performances of our system running on CUDA and MySQL running on a host, but also consider that a relation is too big to be completely put on the DRAM of the display card.