| The existing relational databases, due to its deficiencies in scalability and fault-tolerance, are incapable of fulfilling the requirements of massive data processing in the cloud systems. Consequently, the cloud storage techniques were proposed. However, most of the cloud storage systems, which based on the key-value pairs to store the data, can provide high efficiency for key-based queries. The complex multi-dimensional queries require the scan to the whole datasets, thus, are inefficient in querying. Because the differences among the storage models, the existing index structures are unable to transplant to the cloud storage system. Therefore, the research on the multi-dimensional index structure becomes one of the hot research topics both in the domain of engineering and academic.In this paper, a multi-dimensional index structure, abbreviated as KD-R, was proposed based on the KD tree and R tree. KD-R index, which adopts two layers index model, build the multi-dimensional KD tree index based on the global server while construct the R tree multi-dimensional local index on each local data node. By combing the above two layers' indexing, the scanning of the data node that do not contain the query results can be mostly avoided. In order to optimize the index nodes' publishing strategy from local index to global index, this paper designed a dynamic index adjustment algorithm based on Multi-layer perceptron neural networks. Meanwhile, the KD-R based multidimensional keyword querying algorithm and multidimensional range querying algorithm were proposed. Finally, a series of experiments were illustrated. These experiments were used to test the efficiency of multidimensional quires, the dimension adaptability and the effect of dynamic adjustment of double layer indexing mechanism. The experiment results show that the KD-R index is efficient in improving the performance of multidimensional queries for cloud systems, and it also shows higher extendibility. |
PDF Full Text Request