Thesis Proposal “GPU-accelerated join-order optimization” accepted for publication at the VLDB PhD Workshop
Andreas Meister (University of Magdeburg)
Join-order optimization is an important task during query processing in DBMSs. The execution time of different join orders can vary by several orders of magnitude. Hence, efficient join orders are essential to ensure the efficiency of query processing. Established techniques for join-order optimization pose a challenge for current hardware architectures, because they are mainly sequential algorithms. Current architectures become increasingly heterogeneous by using specialized co-processors such as GPUs. GPUs offer a highly parallel architecture with a higher computational power compared to CPUs. Because join-order optimization benefits from parallel execution, we expect further improvements by using GPUs. Therefore, in this thesis, we adapt join-order optimization approaches to GPUs.