Query Optimization for a Distributed Geographic Information System

Distributed geographic information systems (GISS) have advantages in data sharing, reliability, efficiency, and system growth. Query optimization substantially affects the performance of a distributed GIS. In developing a system, query optimization is one of the technical issues that must first be addressed. A distributed GIs is different from a non-spatial distributed database and requires special techniques for query optimization. In this paper, a set of query optimization techniques are resented that were develo~ed in building a distributed GIs. ' TWO new definitions of spa'tial operations"are introduced that enable us to apply the well-developed operation-ordering approach for &ategy generation. A petrinet-based strat&modeling method is described that is aimed at facilitating " strategy generation and cost estimation. A que$ optimiza2;'on algorithm is presented. Cost functions and selectivityfunctions for spatial operations are described as well. Distributed Geographic information Systems In recent years, distributed GIss have attracted increasing interest. A distributed GIS is a collection of sites connected via a data communication network. Each site is an autonomous GIs that maintains data and processing functions. A distributed GIS provides transparent access to data stored at any of the sites. It presents a single database image and hides data distribution and connection paths. To the user, all the data and functions can be accessed as if they are provided at the local site. Compared with isolated/centralized GIss, distributed GIss have many advantages. The most obvious advantage is the support for data sharing. In many situations, particularly with large data processing projects, data sharing dramatically improves productivity and reduces costs. Additional advantages include improved efficiency, higher reliability, and easier system growth. A distributed GIs can reduce response time. By distributing data properly, the time required for data transmission is minimized. Short response time is also achieved by distributing costly operations to multiple sites for parallel processing. Higher reliability is achieved by duplicating crucial data and functions at multiple sites. In a well-planned system, new computers are easily "plugged in" to incorporate more power. In a word, integrated with data communication networks, GISS may become more accessible, available, and powerful. The advantages and importance of distributed GIss have been realized by GIs researchers and producers (McGregor, 1988; NCGIA, 1989; Meredith, 1995). Some organizational and institutional issues in developing distributed GISS, including the incentives and the impediments, have been addressed by Del~artment of Computing ancl Information Scicnce, University of C;uelpll. Guelph, Ontario N l C 2W1. Canada (fj~~,ang@sno~~l~ite.cis.uoguelph.ca). PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING Iu Wang Pinto and Onsrud (1995). Research has been conducted for developing distributed GISS, for example, by Edmondson (1992), Bernath (1992), Laurini (1993), and Goodman (1994). Recent work includes the DGIS project in Australia (DHPC Project Team, 1996), the DISGIS project in Norway (Norwegian Mapping Authority, 1997), and the geodata modeling technique for distributed GIss at Berkeley (Gardels, 1997). To facilitate geographic data sharing and interoperability, international and national standards have been developed, including the Open Geodata Interoperability Specification (OGIS) (Buehler and McKee, 1996), and the Spatial Archive and Interchange Format (SAIF) (British Columbia Survey and Resource Mapping Branch, 1994). Since 1995, web server-based systems have been developed for geographic data sharing, for example, the Alexandria Digital Library (ADL) (Smith et al., 1996) and many commercial and non-commercial systems (Plewe, 1997). Most of these types of systems support "map-based" queries where a query is used to retrieve geographic data (usually the whole or part of a map) stored at a single remote site. However, progress is slow in building systems that support queries that request map features, evaluate spatial relationships, and involve maps stored at multiple sites. The slow progress may be due, partly, to the special technical problems that must be solved in developing distributed GISs, such as query optimization. Query Optimization Query optimization is the generation of efficient execution strategies for queries. Modern information systems, including advanced GISS, use non-procedural languages to express queries. For a non-procedural query, the system must generate a procedure of operations to execute it. Such a procedure is called a strategy. In a distributed system, the strategy determines the sites and order for executing operations, as well as the procedure for transmitting the requested data. Several strategies may exist for a query, for example, a query requesting data about the regions that have a land cover of "bare soil" and a slope less than five degrees. If the landcover map and the slope map are stored at two different sites and the query is.originated at a third site, we may use at least the following two strategies to obtain the result. The first strategy is to transmit the two maps to the originating site, overlay them and select the result there. The second strategy is to select the regions with the specified cover type or slope at the sites where the maps are stored, transmit the result of one site to the other, overlay the intermediate results and transmit the final result to the originating site. In most situations, the second Photogrammetric Engineering & Remote Sensing Vol. 65, No. 12, December 1999, pp. 1427-1438. 0099-1112/99/6512-1427$3.00/0 O 1999 American Society for Photogrammetry