Simon Riggs <simon@2ndQuadrant.com> writes:
> Right now it seems strange that the index is larger than a btree, yet
> the performance tests show that 3 times as much I/O was used accessing
> the btree.
Well, in an ideal world a hash index probe is O(1) while a btree probe
is O(log N), so that result is exactly what hash proponents would hope
for. Whether it's real or not is another question, but it could be.
regards, tom lane