Folks, I am *not* (not, not, not) talking about "no registration". I am
talking about breaking the deadly assumption that entering a room means
registering with *that server*. This is the MUD model, and it works for
MUDs -- but MUDs are, frankly, toys compared with what we're talking
about building here.
There are at least several critical problems with the space-server-equals-
interaction-server model:
1) Some rooms are going to be enormously popular. Almost no site, no
matter how powerful, is going to be able to cope with 10K people
at once.
2) Some sites aren't going to want to put up interaction servers at all.
That doesn't make them useless. I want to put up some "space" on the
Web, and plan on doing so, but I don't put good odds on talking my
company into shelling out for a machine powerful enough to serve the
interactions as well. That's going to be a *very* common problem; I'd
bet that 3/4 of the people who would put up spaces aren't going to be
able to serve them, at least in the nearish term.
3) Back to example (1) again. Say you've put up the VR equivalent of
CDNow! -- the world's coolest record shop. Thousands of people pile
in. Do you *want* to see them all at once? Think of the *physical*
crowding of the space. You really don't want to deal with more than,
at most, a fraction of the people in that room.
The answer to all of these is to break this unnecessary connection of
"space" and "interaction" servers. The "space" of Cyberspace is *all*
of the Web -- anything connected in. On *top* of that, you have an
unlimited number of servers, each dealing with a collection of *people*,
and the objects that they are coming in contact with. The interaction
servers contain little or no "data" themselves -- they exist solely to
provide the dynamic side of things, essentially deltas to the static
data out on the Web. Obviously, they have some sorts of garbage
collection algorithms, to get rid of the deltas that haven't been
accessed in a long time -- the "janitors" who clean Cyberspace back
up to its native state.
It works, it's *easier* to implement than the world==all servers model,
and it's a helluva lot more scaleable. And it's oriented towards the
notion that a MUD is defined by the *people* in it, not the *space*,
which I think is the case for most good MUDs.
While I'm at it, a side-note: people are assuming that these VRMUDs
are server-centric, just like conventional MUDs, that all data gets
funnelled into the center and redirected back out again. Bad idea;
at the bandwidths we're talking about, that's just begging for
bottlenecks. Instead, assume that the server is just playing traffic
cop, putting the clients in touch with each other and keeping track
of long-term changes to the world. Let the clients do all the hard
work. That's *much* closer to matching the realities of computing
power today...
-- Justin
Who clearly needs to find some time to put
all of this onto his home page somewhere,
so I don't have to keep repeating it...
Random Quote du Jour:
"Questionable: Euphemism employed by regulators and the media to alert
investors to activity which, pending further inquiry, will prove the
perpetrator worthy of stoning."
-- from White Knights and Poison Pills