Message-Id: <9405031543.AA08425@tipper.oit.unc.edu>
To: Christian Huitema <Christian.Huitema@sophia.inria.fr>
Subject: Re: Navigation with centroids
In-Reply-To: Your message of "Tue, 03 May 94 17:32:27 +0200."
<199405031532.AA01029@mitsou.inria.fr>
Date: Tue, 03 May 94 11:43:50 -0400
From: Simon E Spero <ses@tipper.oit.unc.edu>
Each subordinate does not need to keep a centroid of all its parents; it's
just a way of allowing earlier failure detection.
Servers are tried in sequence, but the sequence does not cover all "possible"
servers - just that subtree containing feasible matches, as determined by
centroid based pruning.
The query need not stop at the first match; however the first match is likely
to be the best match, due to the best-first search tree ordering.
At some point it is faster to abandon the localised search, and switch to a
standard authority based resolution strategy. This cut-off point is reached
when the cost of continuing the search is deemed likely to exceed the
maximum savings achievable by using the best copy, as opposed to the
canonical authoritatively mastered copy.
This cut off point varies on the size of the document, and the relative
depths and breadths of the two hierarchies.
Simon