On 10/23/2010 08:54 PM, Tom Lane wrote:
> Another thought here is that the split-in-half rule might be
> unnecessarily dumb. It leaves equal amounts of code space on both sides
> of the new value, even though the odds of subsequent insertions on both
> sides are probably unequal. But I'm not sure if we can predict the
> usage pattern well enough to know which side is more likely.
We can't. In particular, we can't rely on the label to tell us, so we
have no information at all to go on, really. Let's just go with the
simple midpoint.
Are you going to try doing this?
cheers
andrew