Re: 2461bis: address resolution failure as a hard error?

[Erik Nordmark <Erik.Nordmark@xxxxxxx>]

>    The ICMP indication of local address resolution failure SHOULD be 
>    treated as a hard error [RFC1122].
> 
> (the language is a bit wishy-washy because I don't want all the
> address unreachable ICMP messages to be treated like this, just the
> ones which the host in question generated "internally")
> 
> .. the justification is that if one tries to establish communications 
> on a local link, or if the router has gone down and you can't send the 
> packet to anyone, TCP should fail over and try another address (or 
> address family); this is important e.g., if you have v4 and v6 on a 
> link, and they go through different routers.


Two comments:
1. The term "local address resolution" isn't defined so I doubt the 
   above text suggestion clearly communicates what you want to say.
2. I can see this as useful when there is an alternative (which is your 
   motivation), but I'm concerned that treating it as a hard failure when there
   is no alternative is detrimental. 

Another approach would be to attack the more general issue, which I would
state as "when there are alternatives, treat soft errors and an indication
to try an alternative".

   Erik


--------------------------------------------------------------------
IETF IPv6 working group mailing list
ipv6@ietf.org
Administrative Requests: https://www1.ietf.org/mailman/listinfo/ipv6
--------------------------------------------------------------------