A Hybrid Index Structure on Multi-core Cluster Architecture A Job Shop Scheduling Problem in Software Testing A TS-GATS Based Approach for Scheduling Data-intensive Applications in Data Grids An Improved Spectral Clustering Algorithm Based on Random Walk Fairness Analysis of Peer-to-Peer Streaming Systems Image Denoising by 2-D Anisotropic Wavelet Diffusion LogGP（h）： Incorporating Communication Hierarchy into the LogGP Model Neuron Networks Classification Algorithm Based on Bionic Pattern Recognition Optimal Proxy Caching for Peer-to-Peer Assisted Internet On-Demand Video Streaming Services Parallel Sorting for Multisets on Multi-core Computers， Process-level and Thread-level Parallel Programming Mechanism and Performance Optimization Techniques on Multi-core Clusters The Super-node Parallel Systems Based on the Memory Centric Interconnection Webpage Segmentation based on Gomory-Hu Tree Clustering in Undirected Planar Graph
精彩片段：
High-dimensional data indexing and feature based similarity search isemerging as an important search paradigm in computer science. Efficientsupport of them requires power indexing techniques. In this paper wehave proposed an HKD-tree-an efficient parallel algorithm and the par-allel index structure under the SMP cluster architecture to solve the high-dimensional data indexing problem. Our HKD-tree parallel algorithm isbased on the KD-tree and LSH algorithm and outperforms others underthe cluster architecture. A HKD-tree combines positive aspects of bound-ing region based and space partitioning based data structures into a singledata structure to achieve better scalability. It supports queries based onarbitrary distance functions. Our experiments show that a HKD-tree par-allel algorithm is effective support to high-dimensional data spaces andprovides same support of approximate nearest neighbor queries. All in aword， a HKD-tree parallel algorithm and parallel parallel index structurehave excellent performance in SMP cluster architecture. The above exper-iments show that HKD-tree parallel index structure is slightly better thanLSH and KD-tree index structure. It also shows that HKD-tree used inSMP cluster architecture will increase retrieval performance about 30％.So we can know， LSH and KD-tree will be mixed and used in SMP clusterarchitecture will significantly improve the performance of itsalgorithm. As part of our future work， we intend to adjust thread affinity prop-erty of queries like an HKD-treesubtree and the cluster core numberefficiently matching using in parallel structure. We also want to explorethis techniques to support queries in interactive environments efficientlyusing an HKD-tree.
中国大学出版社协会 |
首页 |
宏观指导 |
出版社天地 |
图书代办站 |
教材图书信息 |
教材图书评论 |
在线订购 |
教材征订
