Date: Fri, 23 Jul 1993 23:09:43 +0100 (BST)
From: Martin Hamilton <M.T.Hamilton@lut.ac.uk>
Subject: Re: whois++ // was -- Re: URNs in the DNS
To: Peter Deutsch <peterd@bunyip.com>
In-Reply-To: <9307231702.AA13614@expresso.bunyip.com>
Message-Id: <Pine.3.07.9307232048.B5589-b100000@lust>
Peter Deutsch said:
> As they say on one particularly irritating game show in
> the U.S. "Come on down!" I'd really like to see several
> different servers out ASAP and welcome feedback on what we
> have so far.
Ok! I'm in the midst of writing a particularly boring report
right now, and for the next few weeks will be needing frequent
erm... intellectual stimulation... (or should I say "mental
masturbation" ? Never quite sure about programming :-)
So, expect something along the lines described in the
architecture paper first, with centroids following a bit
later on.
> As for your reservations about centroids, are they
> concerning the specific filtering mechanism, or are they
> concerning the general idea of propagating indexing
> information up multiple trees from the bottom up? I'm not
> convinced myself that the proposed centroids mechanism of
> sending a single copy of all unique strings will be the
> only or most useful indexing scheme used, but I do have a
> lot of faith in the general idea of building such indexing
> trees from the bottom up. Of course, we'll never know
> unless and until we try it.
I can see it working at a specialist level, with individual
index servers (or groups thereof) tracking particular kinds
of resource, but am afraid that connecting these together at
a high level will require all sorts of trickery to avoid
meltdown.
Another gripe is that this strategy sounds like it would work
best when there are a small number of those unique words per
template, but I suspect an index server which is tracking most
of the resources on, say, a big Internet archive will be prone
to meltdown too, given the diversity (no, really!) of the
objects.
Still, there's only one way to find out... !
Cheers,
Martin