%0 Book Section %1 Rauhe2013-he %A Rauhe, Hannes %A Dees, Jonathan %A Sattler, Kai-Uwe %A Faerber, Franz %B Advances in Databases and Information Systems %D 2013 %I Springer Berlin Heidelberg %K All Bulk_synchronous_parallel CPU Dynamic_hashing Efficient_query_execution Expose GPU Hash_table Index_hashing Index_structure Memory_hierarchy Physical_Query_Optimization Query_compilation %P 330--343 %T Multi-level Parallel Query Execution Framework for CPU and GPU %X Recent developments have shown that classic database query execution techniques, such as the iterator model, are no longer optimal to leverage the features of modern hardware architectures. This is especially true for massive parallel architectures, such as many-core processors and GPUs. Here, the processing of single tuples in one step is not enough work to utilize the hardware resources and the cache efficiently and to justify the overhead introduced by iterators. To overcome these problems, we use just-in-time compilation to execute whole OLAP queries on the GPU minimizing the overhead for transfer and synchronization. We describe several patterns, which can be used to build efficient execution plans and achieve the necessary parallelism. Furthermore, we show that we can use similar processing models (and even the same source code) on GPUs and modern CPU architectures, but point out also some differences and limitations for query execution on GPUs. Results from our experimental evaluation using a TPC-H subset show that using these patterns we can achieve a speed-up of up to factor 5 compared to a CPU implementation.

@incollection{Rauhe2013-he, abstract = {Recent developments have shown that classic database query execution techniques, such as the iterator model, are no longer optimal to leverage the features of modern hardware architectures. This is especially true for massive parallel architectures, such as many-core processors and GPUs. Here, the processing of single tuples in one step is not enough work to utilize the hardware resources and the cache efficiently and to justify the overhead introduced by iterators. To overcome these problems, we use just-in-time compilation to execute whole OLAP queries on the GPU minimizing the overhead for transfer and synchronization. We describe several patterns, which can be used to build efficient execution plans and achieve the necessary parallelism. Furthermore, we show that we can use similar processing models (and even the same source code) on GPUs and modern CPU architectures, but point out also some differences and limitations for query execution on GPUs. Results from our experimental evaluation using a TPC-H subset show that using these patterns we can achieve a speed-up of up to factor 5 compared to a CPU implementation.}, added-at = {2015-04-10T18:02:47.000+0200}, author = {Rauhe, Hannes and Dees, Jonathan and Sattler, Kai-Uwe and Faerber, Franz}, biburl = {https://www.bibsonomy.org/bibtex/2c4d0c8cdb30a444ea5f2650a5d81455d/christophv}, booktitle = {Advances in Databases and Information Systems}, interhash = {378e9eaf40ce6c2f3d81458807173f48}, intrahash = {c4d0c8cdb30a444ea5f2650a5d81455d}, keywords = {All Bulk_synchronous_parallel CPU Dynamic_hashing Efficient_query_execution Expose GPU Hash_table Index_hashing Index_structure Memory_hierarchy Physical_Query_Optimization Query_compilation}, pages = {330--343}, publisher = {Springer Berlin Heidelberg}, series = {Lecture Notes in Computer Science}, timestamp = {2016-01-04T14:22:08.000+0100}, title = {Multi-level Parallel Query Execution Framework for {CPU} and {GPU}}, year = 2013 }