Revision 1 - January 2007: In Section 8.3.2, the original report claimed that each step involved four passes over the array, when in fact, it should be five. This means that the figure of 16 passes should be 20, and that the fraction 9/16ths should be 9/20ths. Also, the estimation of two misses per key should be 2.5 misses per key, which agrees exactly with the results in Section 8.3.3. Additionally, the pages were corrupted after a number of graphs. These are fixed in the latest copy of the tech report and code.