Data management on new processors: A survey

Nowadays, data appliances based on the SQL query processing currently face severe problems such as the I/O bottlenecks and high energy consumption when dealing with vast amounts of data. We could not effectively solve these problems by traditional software solutions. A good attempt to mitigate these problems is query off-loading to hardware such as FPGAs or GPUs as a co-processor where partial queries are pushed down to it. Besides increasing the performance, it also reduces the energy consumption, as the hardware serves as the co-processor rather than a CPU. This paper reviews the use of FPGAs and GPUs for query processing in recent years. We first review the challenges and opportunities for data processing, and summarize the use of hardware for data processing. We then highlight several approaches for each component of the query processing, including projections, selections, group by aggregations and joins. Finally, we suggest some future work and summarize our paper.