XP-Trees : External Priority Search Trees
XP-trees (external priority search trees) are simple, practical and versatile structures supporting searches on points, intervals and higher dimensional spatial objects in secondary memory. Our approach to developing external structures is to consider worst-case efficient (internal) data structures from computational geometry, here the priority search tree. With the XP-tree we succeeded in transferring the underlying principle in an appropriate way to organize secondary storage, arriving at a practically useful structure. Together with external counterparts of other structures from computational geometry, such as segment tree, interval tree, and range tree, XP-trees can be used as building blocks to construct nested tree structures directly representing sets of spatial objects in higher dimensions. Like the internal counterpart, the XP-tree supports "halfrange" queries on points in two dimensions. Although XP-trees are not fully dynamic we believe them to be useful in many applications. Regarding balanced XP-trees, O(logdn + t) external accesses in halfrange queries are guaranteed, where n is the number of points, d the degree of the XP-tree, and t the number of points within a halfrange. Mapping intervals, as one-dimensional spatial objects, into two-dimensional points in the standard way, the structure supports all interesting kinds of queries on intervals. An XP-tree also supports queries on spatial objects in more dimensions by projecting them to an interval in one dimension. Hence the XP-tree can be used as an index structure in temporal or geometric databese systems. Experimentalperformance evaluations show that halfrange queries on two-dimensional point sets as well as all types of interval and point queries on sets of intervals are supported efficiently by the structure. Searching is quite fast, when no or few objects are retrieved. Although the shape of an XP-tree degenerates when a point set representation of a set of intervals is stored, the searching behaviour is practically the same as for a corresponding balanced structure. Searching is either very fast or has a large "retrieval ratio".
Nutzung und Vervielfältigung:
Alle Rechte vorbehalten