Index-Partitioning in the Distributed Database System ADAMpro (Bachelor Thesis, Finished)
Author
Description
In order to handle growing collection sizes of multimedia data, the distributed database system ADAMpro offers a variety of index structures which aim at reducing search time. In this thesis, we describe partitioning strategies which can be used to distribute index structures over physical and logical partitions. Our contribution is twofold: On one hand, we outline different strategies to partition data. We describe their properties for usage in large-scale multimedia retrieval and offer a theoretical guideline. On the other hand, we extend the existing system ADAMpro to offer both custom repartitioning of index data and partition skipping at query time with a user-defined granularity. We propose and implement two partitioning schemes which combine some of the outlined ideas. We evaluate on two data sets of one million vectors, one consisting of randomly generated vectors and the other one being the ANN SIFT1M data set and compare to a simple Round Robin partitioning scheme.
Start / End Dates
2016/06/01 - 2016/09/30