RE: PI/metro/geo [Re: The state of IPv6 multihoming development]

["Michel Py" <michel@arneill-py.sacramento.ca.us>]

Aldrin,

> sorry i was referring to IP.  currently, higher-layer
> protocols add reliability and all the other good stuff.
> current multihoming practices (announcing prefixes to
> the internet) try to provide recovery at the IP-layer,
> the time-to-recovery of which widely varies depending
> on the nature of the outage and it's relationship to
> one's site. what should one consider an acceptable
> time-to-recover?  This varies greatly on the kind of
> application one is working with. This is what i meant
> when i said, there really are no layer-3 guarantees
> on the internet today.

We are saying exactly the same thing. Today, a link down will lead to a
BGP reconvergence that will often be longer than a typical TCP
connection will survive. Tomorrow, there is stuff on the table that can
make this delay a lot shorter. If the network guaranteed that the packet
loss would not last more than two seconds, would you say that there is a
layer-3 guarantee on the network?

The point we disagree on is that it's not because it does not exist
today that it means it won't exist tomorrow. If you want more details
email me privately.

Michel.