--- Begin Message ---
On 05/07/2023 16:11, Paul Bone wrote:
Particularly interested in scalable solutions from a few hundred
subscribers up to tens of thousands – but I suspect that may well
involve hardware upgrades to do cost effectively.
(more thoughts)
For just IPv4:
It is just an IPtables rule on whatever terminates sessions (PPP or
vlans or whatever) from your customers? with -s 100.64.0.0/10 ???
Thus you don't end up with 1 massive thing that can fail, and has a very
similar number of single points of failure to actually providing
connectivity? Presuming enough customers sessions per box to reasonably
average it out. Also you will have to be doing some filtering at this
point for to makesure the customer only using IPs they are meant to (BCP38)
And you aren't going to jump to zillions of customers on CGnat on day
1. You will just start dishing out a 100.64.0.0/10 to new customers?
So you can see how the load works out and if necessary add more boxes?
Because if you have $one_massive_blob CG_nat box then you've got to have
all kind of clever (bodge) routing to get to it, and back. And a state
based failover? And 1 massive point of failure.
Tim
--- End Message ---
