The Research And Implement Of Column Database Parallel Query On Heterogeneous Computing Platforms

The database complex query is an important method for information processing andanalysis. With the complexity of applications and development of the Internet, the traditionalline-based database meets some technical bottlenecks. At the same time, the heterogeneouscomputing platform based on multi-core CPU and GPU has powerful computing capabilities,how to use it to exploit the database complex queries efficiency is an important topic indatabase system.In this paper, we has established a new column database parallel query prototype systembased on heterogeneous computing platforms by the steps as:1) Designed a primitive set forcolumn database and translated the SQL into the primitive operations dependency graph.2)Designed a common primitive’s interface, implemented all the primitives on the CPU byusing the page-reading, quick sort, R-B tree and hash table algorithms, and also implementedsome primitives on the GPU by using coalesced, the atomic variables and shared memory.3)Designed a multi-threaded dynamic scheduler to handle the dependence graph. The schedulercan dispatch the primitive operations to be executed on CPU or GPU.4) Used the memorymanager to manage the large number of intermediate vectors.The experiment shows that the database complex queries is a typical compute-intensiveoperations, column databases complex queries has a significant performance than thetraditional line-based database, up to98.62times; the multi-threaded dynamic scheduling caneffectively exploit primitives-level parallelism on the dependency graph, enhance the queryspeed about1.2times; the GPU can exploit the primitive-internal-level parallelism with thespeed up of2.56.In short, this paper appears some effective researchs on how to accelerate databasecomplex query with the heterogeneous computing platforms, established a foundation for adeeper research in future.