Re: ARCH: Virtual spaces and nets

Craig Presson (fcp@moontarz.nuance.com)
Mon, 13 Jun 1994 14:51:48 -0500 (CDT)


> >One project that some people will undoubtedly work on is building
> >virtual copies of real places (cf. Gelerntner, _Mirror Worlds_). For

Oops, it's Gelernter, David H. c. 1991 Oxford U. Press ISBN 0-19-506812-2.

> >this purpose, and for the advantage of familiarity, I recommend we begin
> >with Earth as a shared space and allow multiple coordinate systems,
> >such as lat/long/elev, geocentric Cartesian/polar/spherical, according
> >to usefulness to the given problem.
> >
> >Of course coordinate transformations will be straightforward as long
> >as Cyberspace is kept connected, which it need not be.
>
> I don't think that the coordinate systems need to be decided in advance or
> enforced. One of the advantages of bits over molecules is that they can be
> arranged anyway we want to link them. If I want to build a world where
> rooms are adjacent and don't overlap, that should be fine. I should also
> be allowed to build spaces that do overlap in space and yet I can be in one
> and not the other.

I think it's rather elegant that we can embed metric spaces in
non-metric ones, actually. It would be easy for a given construct's
developer to provide coordinates, and clumsy for the Web as a whole
to do so.

I guess I just did my first flip-flop.

>
> One of the problems of building a very large space is floating point
> precision (ask one of the engineers at Boeing who builds very large
> airplanes with very small parts that must be placed precisely). Anytime we
> have an environment with a large dynamic range, we are going to run into
> problems. With 32 bit floats, I believe that there are about 6 digits of
> precision (help me out on this one if you know exactly). That means that
> objects that are 1,000,000 units apart can only be specified with accuracy
> of 1 unit with respect to each other. In other words, if the UNIVERSE is
> 100,000,000 units across, the local coordinate system of each VRML room
> will center on a 100 x 100 unit grid.

32-bit IEEE-754 has 23-bit fractions, 8-bit exponents, and 1-bit signs,
so you can express .8388608 x 10^255, + or -. So you are about right.
This means you don't set out to model the Universe to arbitrary
precision on a machine that doesn't do DP or EP floating point.

Or build a world-spanning Internet with 32-bit segmented addresses
either :-)

-- fcp@nuance.com (Craig Presson) CPresson@aol.com
President & Principal, T4 Computer Security -------
(205) 880-7692 Voice, -7691 FAX --------------------