Re: Minutes / Notes

[Joe Abley <jabley@isc.org>]

On Monday, Jul 21, 2003, at 02:43 Canada/Eastern, Pekka Savola wrote:

> > Also remember that the effects of a re-homing event
> > can be felt several levels down; if your path to the DFZ is through
> > three intermediate ASes, then you're three-times as likely to feel the
> > impact of a re-homing event if the multihoming architecture presents 
> > an
> > impact to be felt. In some countries, *all* the ISPs are two or three
> > ASes away from the core.
> Nope, it's not as simple as that.
>
> Please remember the context here: we're talking about the connection
> surviability of the end-user.  What happens several AS's upstream 
> doesn't
> typically matter *at all*.  Sure, some circuit goes down, and traffic 
> gets
> re-routed.  These almost always cause a quick re-routing, not e.g. 
> falling
> back to BGP hello timeouts. These are not things that are (typically)
> visible to the end-users, and do not trigger rehoming events.
Right, that's the case right now.

However, if a re-homing event between the user and the DFZ forces a 
renumbering event (for example), and there's no transport-layer support 
for session stability over that renumbering event, then the user will 
see their sessions die.

> The most important things, as it seems to me, are:
>  - your own resiliency
>  - the connection between you and your ISP(s)
>  - your first-hop ISP's network characteristics, configuration, etc.
That is the case with today's IPv4 multi-homing, since the multi-homing 
model provides endpoint address stability. In the case of an alternate 
multi-homing model which doesn't provide address stability, users will 
feel pain.

> So, in this context, I'm not sure that in practice the connection
> survivability is a strict requirement.  IMHO, the quality of your
> first-hop ISP is the thing that matters the most.

Joe