Re: LANG: VRML 1.x Binary Format Proposal

Brian Behlendorf ([email protected])
Wed, 31 May 1995 19:42:48 -0700 (PDT)


On Wed, 31 May 1995, Mark Pesce wrote:
> At 12:49 PM 5/31/95 -0700, Mitra wrote:
> >At 6:55 PM 5/31/95, Mark Waks wrote:
> >>Mitra points out that, for small files, there may be drawbacks to
> >>compression. This may well be true (wouldn't even be surprising), but
> >>is it important? I don't *think* we're talking about replacing the
> >>existing format, just adding alternatives. Am I incorrect? Is anyone
> >>actually proposing that we we move to *only* a binary format? Or, for
> >>that matter, *requiring* compression?
> >
> >No - what I'm trying to point out is that the case of lots of small files
> >may hurt our bandwidth as much as the big files, in which case a binary
> >format would be pretty easy to handle whereas launching gzip costs time and
> >for really small files doesnt gain us much.
>
> It seems reasonable that the compression utility could be embedded into the
> VRML file creator; in a similar manner, the decompression utility could be
> embedded into the VRML file parser. Given this - there's no clear reason to
> use GZIP, is there? - there should be no process overhead associated with
> sending smaller files versus sending larger files.

There is definitely still a benefit to using gzip even if the compression
utility is built into the VRML tools we build. By not marrying the
algorithm to the file format it keeps its highly modular status, and thus
can be an optional function rather than a mandatory one (i.e., people
could still create gzip'd VRML files using the "gzip" utility even if
their VRML tools don't do that on their own).

> I have almost always used Internet over a 14.4 Kbps dialup SLIP (TIA) line -
> as I am now. It has made me acutely conscious of file download times, and
> has thoroughly convinced me that we really do need to attack this problem in
> a substantial manner.

Totally agreed - and the statistical analysis seemed to suggest a gzip'd
ASCII of the average (though well-pruned) VRML file can't be *much*
improved on. Though the book isn't closed, and I encourage other
attempts at finding a tokenization algorithm which could help VRML, I
vote we move on.

Brian

--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=--
[email protected] [email protected] http://www.[hyperreal,organic].com/