Fig 1.
Growth of computing capacity on Intel CPU, Intel Phi co-processor, and NVidia/AMD GPUs.
Fig 2.
Normalized resource use per thread of two different kernels.
Fig 3.
Three different schedules for launching kernels in a hash join between tables R and S.
Fig 4.
Overview of MultiQx-GPU [17].
Fig 5.
Architecture of a typical NVidia GPU [20].
Fig 6.
Two different ways to schedule two CUDA kernels, each of which is in a CUDA stream.
Table 1.
Queries in benchmark.
Fig 7.
Speedup of two-table hash join GPU operations that MultiQx-GPU Optimization and Two-stage Model over sequential solution.
Fig 8.
Percentage of execution time of two table hash join with sequential execution, MultiQx-GPU Optimization and Two-stage Model.
Fig 9.
Speedup of two queries combinations that MultiQx-GPU Optimization and Two-stage Model over sequential solution.
Fig 10.
Speedup of different number of queries that MultiQx-GPU Optimization and Two-stage Model over sequential solution.