Indexing Method for Multidimensional Vector Data

Justin Terry, Bela Stantic

Efficient management of multidimensional data is a challenge when building modern database applications that involve many fold data such as temporal, spatial, data warehousing, bio-informatics, etc. This problem stems from the fact that multidimensional data has no order that preserves proximity. The majority of existing solutions to this problem cannot be easily integrated into the current relational database systems since they require modifications to the kernel. A prominent class of methods that can use existing access structures are ‘space filling curves’. In this work we describe a method that is also based on the space filling curve approach, but in contrast to earlier methods, it connects regions of various sizes rather than points in multidimensional space. Our approach allows efficient transformation of interval queries into regions of data which results in significant improvements when accessing the data. In detailed empirical study, we have demonstrated that the proposed method, which can be integrated within the commercial RDBMS, outperforms the best available off-the-shelf methods for accessing multidimensional point data.