No abstract

Processor manufacturers build increasingly specialized processors to mitigate the effects of the power wall to deliver improved performance. Currently, database engines are manually optimized for each processor: A costly and error prone process.In this paper, we propose concepts to enable the database engine to perform per-processor optimization automatically. Our core idea is to create variants of generated code and to learn a fast variant for each processor. We create variants by modifying parallelization strategies, specializing data structures, and applying different code transformations.Our experimental results show that the performance of variants may diverge up to two orders of magnitude. Therefore, we need to generate custom code for each processor to achieve peak performance. We show that our approach finds a fast custom variant for multi-core CPUs, GPUs, and MICs.

No abstract

Graphics processing units (GPUs) promise spectacular performance advantages when used as database coprocessors. Their massive compute capacity, however, is often hampered by control flow divergence caused by non-uniform data distributions. When data-parallel work items demand for different amounts or types of processing, instructions execute with lowered efficiency. Query compilation techniques---a recent advance in GPU-accelerated database processing---suffer from the problem even more, because divergence effects are amplified during the execution of fused pipeline operators. In this work, we identify two types of control flow divergence--- filter divergence and expansion divergence ---that frequently occur in real world workloads. We quantify the problem for two poster cases and propose techniques to balance these divergence effects. By balancing divergence effects, our approach is able to restore processing efficiency even when pipelines contain heavily skewed operations. Our query compiler DogQC has a wider range of functionality than other query coprocessors and achieves performance improvements. We observe shorter execution times for TPC-H benchmark queries by factors up to 4.51x compared with existing GPU query compilers and by factors up to 4.54x compared with CPU-based systems.

No abstract