The Sequoia data that was used in the paper, did not include all of the polygons and islands from the Sequoia data set. In particular, some of the very large polygon attributes (polygon attributes that required more than 8K to represent on disk), were omitted. These large polygons ended up joining with many small islands, and as a result, the time for performing the join was dominated by the time to write the result relation to the disk. To limit the size of the result, one could maintain a single copy of the large polygon, and in the result tuples that refer to the large polygon, point to that single copy. The actual Sequoia data size is as follows: Polygon: 58586 objects, relation size: 27.4 MB Islands: 21021 objects, relation size: 5.8 MB The Result of the join has 35767 tuples and is 254.6 MB in size. The time to perform the spatial join using PBSM and a 8MB of buffer pool, is 4330.7 seconds. The time for the join is largely dominated by the refinement step which takes 4203.1 seconds. The Paradise query to perform the join was: "select * from polygon, islands where shape >= hole" In Paradise SQL, ">=" between two polyons is overloaded to imply containment.