[Dundi] Dundi w/ 1.0

[steve at daviesfam.org]

On Wed, 20 Oct 2004, Mark Spencer wrote:

> Okay so to illustrate...
> 
> The core tier 1 people will all be exchanging *most* call setups.  The 
> challenge is to try to trim off as many branches as possible and to reduce 
> as much as possible the number of queries that have to go deep into the 
> tree.  To this end, it is beneficial to have numbers which are similar 
> (e.g. all Netherlands numbers) behind a one or more tier 1 / tier 2 nodes. 
> Note that the protocol is designed to handle *any* graph and any 
> distribution of numbers, but certain orientations are more effective in 
> the value they bring to the cache and in their ability to block off large 
> chunks of the tree.
> 
> Mark


An observation - we are used to the numbering plan being geographically
heirarchical.  We tend to unconciously think that it "has to be this way".

People are talking about local number portability, but lets think about a
world where numbers aren't geographically arranged at all.  I might
publicise a "number" (steve at daviesfam.org anyone?  GET-DIGIUM-IN-ZA?  
Best-bananas-$1-a-kilogram?  911(!)) which says nothing about where it is
located.

The neat thing about Dundi and the Dundi e164 "network" is that it can
work nearly as well in this world as in the current world.  (It does of 
course reduce the ability of the current implementation to prune the 
tree, we should consider a "dns" and or "email" context with appropriate 
pruning rules)

All that is necessary is that the paths into the tree from any point need
to go from "more likely to call" to "less likely to call".

Generally people call others close by more than they call people far away.  
So it would be probably be a good idea to peer with those close by - like
in my city.  Who, in turn, will peer with others relatively speaking close
by - like in their state.  etc etc.
  
Perhaps, you also call others in an "interest group" - Asterisk geeks, for
instance.  In which case, you may want to peer with a peer in that
community.  Now we see the tree morphing away from purely geographic, but 
reflecting the real communities of calling.

If people follow this natural flow then our query tree will be efficient.  
It'll reflect the actual "closeness" of people in an abstracted kind of 
way.

It's a bit like if you attempted to map the world using the BGP mesh.  
You'd find, for example, that the UK and US are much closer together than
the UK and Iceland.

Once the network gets larger, frequent practical examination of the flow
of queries will be necessary in order to identify inefficiencies and
adjust peerings.

But people are people.  My first reaction was "I wanna be on the list of 
uber-1st-tier peers"!  I wanna peer with the big boys!  Just like everyone 
wants to take their NTP off a Stratum-1 server.

But really I need to think: who do I call most?  Who is most likely to
call me?  Those are the places I should peer into.

Or at least that's how it looks to me.

Regards,
Steve