Damn it, now we're going to have to make hashing faster
Conversation
Replying to
hash_mem_multiplier's default was increased to 2.0 in 15, so that's something! Though it's equally relevant to hashing and sorting.
Maybe the cost model of sorting should be adjusted now that it has become more efficient. The constant factor should depend on the sort input type (compare cost) and row size (swap cost).
1
That's probably a good idea. IIRC the other problem with cost_sort() is that the CPU cost of external sorts is too pessimistic.
1
Show replies