Alternative strategies for Performing Spatial Joins on Web Sources

Abstract.

With the current information explosion on the Web, numerous applications require access to a collection of different but related pieces of distributed geospatial data. In this paper, we focus on one set of such applications that requires efficient support of spatial operations (specifically, spatial join) on distributed non-database sources. The main challenge with this environment is that remote sources are usually read-only and/or do not support spatial queries. Moreover, several of these Web-based applications can tolerate either some level of inaccuracy or progressively filtered (or polished) results. Therefore, conventional distributed spatial join strategies are not applicable or efficient in this environment. To address these challenges, we first break down the process of distributed spatial join operation into three steps: (1) local to remote transfer, (2) remote spatial selection, and (3) local refinement. Then, for each step, we propose and study alternative techniques and by varying their combinations, we generate several query plans. Each plan strives to strike a compromise between efficiency and accuracy. Since the techniques proposed for the first step have significant impact on the overall performance of the query, we specially focus our attention on this step. We propose two heuristics for the first step to reduce either the number of selection queries or the area covered by each selection query. Within a realistic experimental set-up, we show that one heuristic is more appropriate with fast networks and a powerful local server, while the other one is superior in the opposite situation. Our experiments also show that both heuristics outperform approaches based on transmitting either the actual spatial objects or their bounding boxes. Note that the intention of this paper is not to propose a query optimizer to choose one plan over the others. Instead, it serves as a first step towards the design of such an optimizer by concentrating on the design and evaluation of several alternative plans within a realistic experimental set-up.