[ogpx] Globally unique authentication token versus Firstname Lastname [Was: Re: Global Agent Domain Registry [Was: Re: OpenID and OGP : beginning the discussion ...]

[Carlo Wood <[e-mail]>]

On Sun, Jun 28, 2009 at 09:06:31AM -0700, Infinity Linden wrote:
> okay... so it seems like you're actually asking... "how do we maintain
> global avatar name uniqueness if avatar first and last name are used
> as primary keys for logging in?"

Yes. Because, it *WILL* become a problem if two people can use (steal)
the name of someone else. And therefore we need to garantee uniqueness
of the identifier that is visible even (the tag!), and also, we need
one central database to garantee this uniqueness if the keys itself
ARE the first+last name.

This is a very serious problem.

> there was some discussion about a global avatar name registry last
> year in the SL Architecture Working Group, though i don't think there
> were any definitive results.
> 
> i think we're punting on the problem with respect to protocol
> development. that is, i don't think the protocol specifications will
> say anything about the expectation that an agent name be globally
> unique across grids.

Well, it will have to; I am convinced that if we ignore this
problem now, it will turn out to require such a drastic protocol
change in the future that you can never solve it anymore (ake, IRC).

> deploying such a service would be quite an effort in terms of community buy-in.
> 
> that being said, there's no reason it can't be done in the future or
> as a project of a non-IETF group.
> 
> for the time being, we can use account identifiers to guarantee
> uniqueness; or at least that's my sense given the lack of radical
> agreement in the topic.

I wrote another article about this topic (again, for IRC) in the past;
that is - I have already put a lot of thinking into this subject.
I don't think ignoring the problem is going to make it go away.

> in other words, i think it's a topic that will lead to wailing and
> gnashing of teeth and would only serve to delay production of
> standards that people _do_ agree on.

Not at all necessary, I have a solution :)

But, like the protocol negotiation one, it turns out that most
people are not ready to grasp the concepts involved and rather
just ignore it :/

Ok,... here it goes, I'll try to be extremely short, which
won't make it more understandable, but please ask any questions
until things are completely clear.
----------------------------------

* Uniqueness

Some token can ONLY be unique if it is *generated* by an
authority. Unless we use ONE global authority, the generated
token must include a sub-token that was assigned to that
particular authority that is globally unique for that authority.

Example; say there are three authorities. Each get assigned
a unique token: A, B and C. Upon request they generate a
locally unique token (for example an integer that is incremented
every time) and prepend their own token to make it globally
unique (ie, A1, A2, .. B1, B2, .. etc).

* Human readable identifiers

Because the authentication token have to be generated, they
cannot be human readable without creating a major problem.
I am convinced that it is near impossible to generate globally
unique and human readable Firstname Lastname for -say-
100,000,000 people, without getting such long and unreadable
last names that impersonations are still possible (in the least!)

Hence, the human readable identifier will have to be separated
from the really (globally) unique authentication token.

The reason that this can work in most cases is simply because
everyone only meets a limited number of people in their lives.
This number, especially if you don't count people that just
walk one once, but only look at people you interact with,
is almost a constant: you cannot have infinite number of friends.
For example, every user will have to do with at most 1000 others,
independent of the total number of users of all virtual worlds
together.

* Solving namespace collisions

It will happen (by coincidence or deliberate) that two people
with the same name will run into eachother and/or have interactions
with the same people. How to solve this?

My solution for this is as follows:

Each agent has separated their authentication token (something like
a UUID, only used internally) and their human readable "tag".
The "tag" exists of a Firstname, a Lastname and a postfix.

If an agent meets someone for the first time, it will receive
their authentication token. The user (the viewer) will store
this authentication token along with the a chosen postfix.
If the Firstname Lastname of the new avatar is unique within
the scope of the viewer, it will automatically assign an
empty postfix (although the user is allowed to edit it).
If another authentication token already exists with the same
Firstname Lastname, then the viewer will warn the user and
require him to choose a (locally) unique postfix. Because
the user can choose, it will be not only human readable but
also easy to remember.

For example, I know a Paul Mastermind that looks like kid
and is my friend. Must later I run into an other avatar
with the same name Paul Mastermind; my viewer assigns a
postfix '2' to it, so that I see in the tag:
"Paul Mastermind 2". Of course, this Paul looks totally
different: it's a man that can play chess very well, and
he offers me to teach me chess. I edit his postfix and
change it into "Chess master", so now I see:
"Paul Mastermind Chess master". My viewer accepts this
because it is unique within the two Paul Mastermind's
that is known. Later I meet a Paul Mastermind 2 that
looks exactly like the Paul Mastermind that I know.
I talk with him and he convinces me quickly that it's
an ALT of the real Paul Mastermind. I edit his postfix
and change the '2' into "ALT".


The point being: the protocol needs a DRASTICAL change
for this to be possible that we need to take into account
NOW. Namely: everyone should be identified by some separate,
and in practise impossibly human readable indentifier.
The protocol can be designed by just saying that this
identifier exists and is some string, and we'll take
care of it's uniqueness later. But really, it has to be
something different from the human readable Firstname Lastname.

Therefore we cannot postpone this discussion.

-- 
Carlo Wood <[e-mail]>