[Dundi] advertising for peers

Mark Spencer markster at digium.com
Fri Oct 22 15:23:59 CDT 2004


> The peering graph ideally wants to map to the number space, and not to 
> anything else. Suppose one organisation is able to terminate calls matching 
> _+1650. and _+612. and _+1519. -- should that organisation be trying to peer 
> three times, once per corresponding branch?

It would be more optimal if that were the case.  Again, it's not a 
requirement, just an optimization.

Suppose for example, you were in the U.S. and represented 100 area codes. 
You would still only receive no more than 900 calls per cache-time-unit 
that would be for area codes you didn't represent, and probably much less.

> Bigger, nexus peers are going to be able to match a large range of numbers, 
> but there are fewer of them. Smaller, itty-bitty enterprise peers are going 
> to be able to match a much smaller range of numbers, but there will be 
> potentially many more of them.

Right, and to the degree that those caches can be maintained higher up we 
should try to do that.

> Is it more important for the big peers to connect to branches in the right 
> place, or the smaller peers, in terms of scaling the system?

Actually it's more the smaller peers because if a smaller peer adds an 
area code, it affects the entire tree in which that smaller peer lives. 
Hopefully, the statistics I mentioned in my roadmap will help us to 
identify them (since in principle the cache hit performance will be 
lower).

Mark


More information about the Dundi mailing list