markus@bluegap.ch (Markus Schiltknecht) writes:
> Simon Riggs wrote:
>> With that in mind, can I clarify what you're thinking, please?
>
> Sure, I can try to clarify:
>
>> 2) the things you've been discussing are essential requirements of
>> partitioning and we could never consider it complete until they are also
>> included and we must therefore talk about them now to check that its all
>> possible before we do anything on SE
>
> I thought so, but am slowly dropping that point of view. In favor of
> something like: hey, if you manage to do it all automatically, cool,
> go for it!
>
>> 3) doing SE first is right, I'm just thinking ahead
>
> Yes, SE certainly has merit. Combine it with some sort of maintained
> CLUSTERing order and it's worth doing, IMO.
My suspicion is that if this gets added in with "maintained CLUSTER
order," we *lose* all of the exclusions aside from the ones directly
established by the CLUSTER order.
That is, the CLUSTER ordering winds up preventing other "natural"
patterns from emerging, with the result that SE winds up being of
pretty limited usefulness.
> I'm not convinced about dynamic partitioning being able to generally
> replace explicit partitioning anytime soon.
It also seems to me that explicit partitioning would make this form of
dynamic partitioning less useful.
Suppose there are 4 more or less uniformly used partitions; if you're
splitting the data evenly across 4x the partitions, then that means
that each segment will tend to have ranges ~4x as wide, which makes SE
rather less of a "win." (Relax the assumption of uniform
distributions, and that just changes the weights...)
--
output = reverse("ofni.sesabatadxunil" "@" "enworbbc")
http://cbbrowne.com/info/advocacy.html
"It seems that perfection is attained not when nothing is left to add,
but when nothing is left to be taken away."
-- Antoine de Saint-Exupery.