[RFC] Strawman replacement local directory state
Michael Ellerman
michael at ellerman.id.au
Mon Jun 19 01:13:29 BST 2006
On 6/19/06, John Arbash Meinel <john at arbash-meinel.com> wrote:
> Michael Ellerman wrote:
>
> ...
>
> >> >
> >> > That's not the point though. A data structure like this makes it
> >> > impossible/hard to speedup partial-tree operations.
> >>
> >> Not at all. Its structured specifically to allow random access (with the
> >> dual delimiters in place, this is now ready for use)
> >
> > OK, your initial description was "read all/write all" I think which I
> > thought precluded random access.
> >
> > cheers
>
> Well, there were versions where it would have been difficult (though
> probably never impossible). But over the course of evolving it, we have
> actually made it pretty easy. And I worked out a way for it to be
> reasonable without actually affecting performance.
Right, from your experiments it does sound like a pretty cool format.
> I welcome the fact that you bring up big big N scaling, so that we can
> make sure to have a reasonable answer. Even if it isn't implemented yet,
> we should have a way to scale (once we really understand the problems
> involved).
Cool. I think any medium-term data format needs to at least think
about scaling to the next order of magnitude or two from the current
limits. And yeah, it doesn't matter if the current implementation
doesn't scale, it's the potential scalability of the format that
matters.
cheers
More information about the bazaar
mailing list