30.01.2014 Views

Annual Report 2010 - Fachgruppe Informatik an der RWTH Aachen ...

Annual Report 2010 - Fachgruppe Informatik an der RWTH Aachen ...

Annual Report 2010 - Fachgruppe Informatik an der RWTH Aachen ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Fast Access to Spatial-Temporal Data<br />

High Dimensional Indexing<br />

Steph<strong>an</strong> Günnem<strong>an</strong>n, Hardy Kremer<br />

Recent applications dem<strong>an</strong>d fast query response times on high dimensional data. For this<br />

purpose index structures were introduced. Existing multidimensional indexes like the R-tree<br />

provide efficient querying for only relatively few dimensions. Therefore we develop new<br />

index structures for efficient retrieval <strong>an</strong>d similarity search.<br />

Due to massive overlap of index descriptors, multidimensional indexes degenerate for high<br />

dimensions <strong>an</strong>d access the entire data by r<strong>an</strong>dom I/O. Consequently, the efficiency benefits of<br />

indexing are lost. By exploiting inherent properties of the indexed data, our new index<br />

structures, the TS-Tree <strong>an</strong>d the OF-Tree, c<strong>an</strong> index high-dimensional data in <strong>an</strong> overlap-free<br />

m<strong>an</strong>ner; during query processing, powerful pruning via qu<strong>an</strong>tized separator <strong>an</strong>d metadata<br />

information greatly reduces the number of pages which have to be accessed, resulting in<br />

subst<strong>an</strong>tial speed-up.<br />

Due to the increasing main memory capacity of mo<strong>der</strong>n computers, a high percentage of<br />

datasets fits into main memory. We develop novel main memory based index structures that<br />

use individual dimensions for each data object by applying the method of subspace clustering.<br />

By a local selection of dimensions we increase the information content for objects compared<br />

to a global approach; this higher information content enables a better pruning of the search<br />

space.<br />

Figure 12: Representation in data space (left) <strong>an</strong>d data structure (right) of <strong>an</strong> overlap free R-tree.<br />

314

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!