The Impact of Buffering on Closest Pairs Queries Using R-Trees

In this paper, the most appropriate buffer structure, page replacement policy and buffering scheme for closest pairs queries, where both spatial datasets are stored in R-trees, are investigated. Three buffer structures (i.e. single, hybrid and by levels) over two buffering schemes (i.e. local to each R-tree, and global to the query) using several page replacement algorithms (e.g. FIFO, LRU, 2Q, etc.) are studied. In order to answer K closest pair queries (K-CPQs, with K ≥ 1) we employ recursive and non-recursive (iterative) branch-and-bound algorithms. The outcome of this study is the derivation of the outperforming configuration (in terms of buffer structure, page replacement algorithm and buffering scheme) for CPQs. In all cases, the savings in disk accesses is larger for a recursive algorithm than for a non-recursive one, in the presence of buffer space. Also, the global buffering scheme is more appropriate for small or medium buffer sizes for recursive algorithms, whereas the local scheme is the best choice for large buffers. If we use non-recursive algorithms, the global buffering scheme is the best choice in all cases. Moreover, LRU is the most appropriate page replacement algorithm for small or medium buffer sizes for both types of branch-and-bound algorithms. FIFO and LRU are the best choices for recursive algorithms and 2Q for the non-recursive ones, when the buffer is large enough.