... description of two desired caching schemes-- exact models (URL:
http://foo.bleuch.com/some/path/chair.vrml), and inexact models ("just
give me a chair") deleted:
> The question is, how do we support both of these models? The answer,
> I believe, is keywords.
>
> When we specify an object to be fetched in VRML, we may specify one
or
> two related lists of keywords. One would be a list of keywords that
an
> object *must* satisfy if it is to be considered acceptable for this
> use: "We need a six-foot, left-handed fridge". Another would be a
list
> of optional keywords, in priority order; the user's Browser would be
> responsible for deciding whether to fetch the exact object from the
> scene's machine, or to use a close local approximation.
> Critique?
Caching things you've seen before is easy; you just add an entry to a
dictionary that maps from URL to Object * whenever you grab a URL from
the net, and check the dictionary before grabbing a URL off the net.
You could do the same thing except use keywords, but frankly, I don't
see the point. What are the chances that I've run across a "six-foot,
left-handed fridge" since I started up Mosaic (or since Mosaic ran out
of memory in its cache and had to flush it-- remember, we're talking
about this exactly because we think these objects will be large)?
Unless we assume that most VRML scenes look about the same, with about
the same set of objects, the caching by keywords won't work much better
than caching by URL.
If we do assume that most VRML scenes look about the same, then count
me out; I have a short attention span, and will get bored.
If there was some sort of database that indexed into objects stored on
the local disk, that database could be loaded at startup time and the
local filesystem could be used as the object cache. I'm going to claim
there's no way that is going to work, either-- I'm not going to
dedicate space on my disk to VRML objects, and I'm not going to
dedicate the effort required to keep the keyword index database
up-to-date.
I see all of this discussion as completely orthogonal to VRML. I
believe it is already being hashed to death by the folks talking about
URI (uniform resource indicators), where the idea is to distribute
commonly used information to many places across the web and refer to it
in such a way that the nearest server could be used to fetch it.