North American Network Operators Group Date Prev | Date Next | Date Index | Thread Index | Author Index | Historical Re: hole punching reality check
In message <[email protected]>, Randy Bush writes: > Provider P allocates a block of address space A to customer, I an ISP. > > I allocates a /24, call it B, out of that block to one of their customers, C. > > C later leaves ISP I and gets a wire directly from provider P. > > I wishes to give C time to renumber. > > I would expect the world to see the following announcements. > > A - I announcing a shorter prefix, A > B - P announcing a more specific for the /24 allocated to C > > Is this not what we all expect? > > Why I ask is we have provider P telling I that they should stop announcing A > and instead announce a bunch of prefixes around the /24. > > randy Let's get the whole picture: before the move: P announces A'/L1 to <the world> B announces A/L2 to P (where L2 > L1) C announces B/24 to B There is no reason for P to announce anything more specific than A'/L1 to the rest of the world unless B is multihomed. If there is no A'/L1 aggregate then never mind. after the move (if B is single homed to P): P announces A'/L1 to <the world> B announces A/L2 to P (where L2 > L1, L2 < 24) C announces B/24 to B There is no need for P to announce anything more specific than A'/L1 if B is not multihomed. If A'/L1 doesn't exist, there is still no reason to announce anything more specific than A/L2. if B is multihomed: P announces A'/L1 to <the world> B announces A/L2 to P (where L2 > L1, L2 < 24) B announces A/L2 to <other provider O> P announces A/L2 to <the world> O announces A/L2 to <the world> C announces B/24 to B P announces B/24 to <the world> The last part is unfortunate. Curtis - - - - - - - - - - - - - - - - -
|