Declustering Spatial Objects by Clustering for Parallel Disks

In this paper, we propose an eÆcient declustering algorithm which is adaptable in di erent data distribution. Previous declustering algorithms have a potential drawback by assuming data distribution is uniform. However, our method shows a good declustering performance for spatial data regardless of data distribution by taking it into consideration. First, we apply a spatial clustering algorithm to nd the distribution in the underlying data and then allocate a disk page to each unit of cluster. Second, we analyize the e ect of outliers on the performance of declustering algorithm and propose to handle them separately. Experimental results show that these approaches outperform traditional declustering algorithms based on tiling and mapping function such as DM, FX, HCAM and Golden Ratio Sequence.