National Conference on Advances in Computing |
Foundation of Computer Science USA |
NCAC2015 - Number 7 |
December 2015 |
Authors: Pramod B. Deshmukh, Yogesh B. Lokare, Ajay V. Katware, Pankaj A. Patil |
dcb8ff7f-6398-4270-9719-51599f5c86b3 |
Pramod B. Deshmukh, Yogesh B. Lokare, Ajay V. Katware, Pankaj A. Patil . A Survey on Massively Parallelism for Indexing Multidimensional Datasets on the GPU. National Conference on Advances in Computing. NCAC2015, 7 (December 2015), 18-24.
CUDA is a parallel programming environment that facilitates significant performance improvement by leveraging the massively parallel processing capability of the GPU. The general purpose computing, on graphics processing unit (GP-GPU) has turn up as a new cost effective parallel computing framework, in high performance computing research that enables large amounts of datasets to be processed in parallel. Large scale scientific data intensive applications have been playing a major role in modern high performance computing research. This large amount of data can be accessed by scientific data analysis applications such as multi-dimensional range query, but not much research has been conducted on multidimensional range query on the GP-GPU. Inherently multi-dimensional indexing trees such as R-Trees are not well suited for GPU environment because of its irregular tree traversal. It has been known that traversing hierarchical tree structures in an irregular manner make it difficult to exploit parallelism and to maximize the utilization of GPU processing units. Then to avoid the drawbacks of R-Tree the novel MPTS (Massively Parallel Three-phase Scanning) R-tree traversal algorithm for multi-dimensional range query was proposed, that Recursive access to tree nodes into sequential access. Furthermore, the recursive tree search algorithms often fail because of the GPU's tiny runtime stack size. Then the proposed work of a novel parallel tree traversal algorithmâmassively parallel restart scanning (MPRS) for multi-dimensional range queries avoids recursion and irregular memory access. Then the proposed MPRS algorithm traverses hierarchical tree structures with mostly contiguous memory access patterns without recursion, which offers more chances to optimize the parallel SIMD algorithm