Martijn van Oosterhout <kleptog@svana.org> writes:
> The cost on inserting would generally go to either penalty, or
> picksplit. Certainly if you're inserting lots of values in a short
> interval, I can imagine picksplit being nasty, since the algorithms for
> a lot of datatypes are not really reknown for their speed.
Tut tut ... in the words of the sage, it is a capital mistake to
theorize in advance of the data. You may well be right, but on the
other hand it could easily be something dumb like an O(N^2) loop over
a list structure.
I'll post some gprof numbers after Stephan sends me the dump. We
should probably move the thread to someplace like pgsql-perform, too.
regards, tom lane