RE: LANG: Binary formats.

Linas Vepstas (linas@innerdoor.austin.ibm.com)
Thu, 9 Mar 1995 17:43:35 -0600


> From: Jan Hardenbergh <jch@nell.oki.com>
> To: www-vrml <www-vrml@wired.com>
> Subject: RE: LANG: Binary formats.
> Date: Tue, 28 Feb 95 14:53:00 E
>
>
> It might well be that the best thing we can do for the data transfer is to
> compress the files. gzip (gnu zip) compresses the Connaly molecule from
> 635K to 181K.
>
> The text data usually about 8 chars per floating point value. A binary
> representation is 4 bytes. That is only a factor of 2. A binary version
> of the Connaly molecule is about 235K. (using Inventor on NT)

Wow. I think this is an important observation that argues heavily against
binary formats. A couple smaller issues remain:

a) Q: Why not have binary & compress? A: I have many files containing binary
floating pt. data. Many of these are incompressible by either compress
or gzip. The ones that do compress don't shrink by a lot (10% to 30%)
Maybe these are especially nasty files ... but ... maybe they are typical.
Can you gzip the binary conelley molecule?

*) Q: Should vrml server/client compress/decompress, or should the lan/wan?
I note that many modems, 14.4 & ISDN, offer compression ... which works
for text, but not for gif's. Once something is compressed, it can't be
compressed further. So shipping compressed vrml over an isdn modem only
ends up chewing through more client-cpu time, without offering any actual
transmission benefit. So maybe vrml should punt, and let the network
equipment do its job?

--linas

Sincerely,
Linas Vepstas

^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^
<a href="http://w3.austin.ibm.com/~linas">
Linas Vepstas Graphics Architecture
IBM PPS Zip 9260 Tie Line: 678-1116
11400 Burnet Road External Phone: 1-(512)-838-1116
Austin TX 78756-9260 mail: linas@austin.ibm.com </a>
^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^v^