[arin-ppml] Rationale for /22
owen at delong.com
Wed Jul 29 11:20:18 EDT 2009
On Jul 28, 2009, at 9:48 PM, Jon Lewis wrote:
> On Tue, 28 Jul 2009, William Herrin wrote:
>> In that scenario, nothing but some inefficient routing. Now let's
>> suppose you're connected to Sprint and Sprint and XO have a peering
>> spat. You should still be able to get to me since we're both
>> to Sprint but you can't because you filter my /24 route and XO's
>> covering route has vanished from your view.
>> Sprint and Cogent had that problem last year. It lasted a good part
>> of a week.
> This was originally going to be my reply to Owen, but then I noticed
> his reply was not to the list.
You're welcome to include what I sent you to the list. I sent it to
you because I did not
expect it to be of general interest to the list, not because I cared
about it being private.
> Ok, there are a few assumptions in the scenario I suggested.
> 1) "Tier 1's" "don't filter"...i.e. not for route table reduction.
While this is true of ~90% of them today, I think as the routing table
expands in the IPv4
end-game, it will rapidly become progressively less true.
> 2) Any network that does filter for the purpose of route reduction
> carries a default route pointing towards a network that doesn't
> doesn't filter or that points default at one that doesn't. This
> could go on for several AS's.
Relying on this fact assumes an ability to detect changes in behavior
AS's away and respond to them in a timely fashion or risk rather
difficult to debug problems
down the road.
> 3) For the sake of this argument, peering battles are ignored. Yes,
> they happen from time to time, and they're one of the reasons
> networks multihome, but they're hardly the normal state.
True, but, when they do occur, your recommended behavior will
exacerbate the consequences
to end users.
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 2105 bytes
Desc: not available
More information about the ARIN-PPML