Gipsy

TRANSFORMERS: Robust spatial joins on non-uniform data distributions

“…We measure the execution time (without taking into account the indexing phase) for the join for the fastest and most broadly used disk-based spatial join methods, i.e., PBSM [3], the Synchronized R-Tree (R-Tree [2]) and GIPSY [4]. We use the best configuration for each approach, e.g., the number of partitions/tiles for PBSM, page size and fanout for the R-Tree.…”

“…As Figure 1 shows, doing so proves efficient on contrasting densities but their inherent problem of structural overlap leads them to read and test more data than necessary, making them comparatively slow when joining similar densities. GIPSY [4] is based on fine-grained, data-oriented partitioning and on neighborhood information. It minimizes the impact of overlap by using the sparse dataset to selectively retrieve the data needed from the dense dataset, relying on connectivity information instead of a hierarchical tree traversal.…”

“…The GIPSY [4] spatial join uses data-oriented partitioning but minimizes the impact of overlap by using a crawling strategy [8], [9]. GIPSY is efficient for joining a sparse dataset with a dense one.…”

“…Methods developed in the past can efficiently join two or more disk-based spatial datasets of uniform distribution of element locations. They are, however, inadequate when joining two spatial datasets, each with a skewed distribution of element location [4]. Doing so, however, is important as the datasets joined rarely have a similar distribution.…”

“…Approaches based on space-oriented partitioning (e.g., PBSM [3]) do well when joining datasets of similar density while dataoriented partitioning approaches (e.g., based on the R-Tree [2], [5] and particularly GIPSY [4]) are more efficient when joining datasets with contrasting density. None of the existing approaches joins the datasets (or areas) with robust performance across different density ratios.…”