Re: draft-baker-6man-multiprefix-default-route-00.txt is a newdraft

Hi Albert,

El 12/11/2007, a las 19:28, Manfredi, Albert E escribió:

>
> I found nothing objectionable at all in the draft.
>
> Matter of fact, it seems to address something that also occurs with
> IPv4, with multihomed hosts. And that apparently, some OSs screw up
> royally. Which is, if a multi-homed IPv4 host, connected to two
> different IP subnets, transmits an IP packet over Subnet A, it  
> often is
> found to use as source address the address assigned to its Subnet B
> interface.
>

Please note that the situation that this draft addresses is somehow  
different than the one you mention, since both addresses (with  
possible different prefixes) can be assigned to the same interface

So if different interfaces have different addresses and when the host  
sends a packet through a given interface using the address of the  
other one as source address, this is easier to solve, since you only  
need to honor the address assigned to the outgoing interface

However when both addresses are assigned to the same interface, it is  
harder to do the selection and it is likely that additional  
(external) information is needed e.g. the next hop router or the exit  
ISP), which is a harder problem imho, and i guess this is what Fred's  
draft addresses

THe result is that you end up routing packets taking into account  
their source address, which is different than what is done today.

This basicallly means that for some packets (packet with external  
destiantion address) the exit ISP selection is performed by the hosts  
and NOT by the intra site routing system. Moreover, failures will  
need to be taken care of by the hosts itself, bascially retrying with  
a different source address

While i think all this is very good, i think it is imporntat to state  
up fron the concequences of this change and compare them with current  
practices.

I certainly agree with Fred with the fact that if you don't do  
something like this, then you rpobably will have to drop packets due  
to ingress filtering incopatibility

Regards, marcelo

> Whether or not this behavior results in total comm failure does not
> preclude that is seems just plain wrong.
>
> I agree that calling this "source routing." or anything similar, would
> be misleading.
>
> Bert
>
> --------------------------------------------------------------------
> IETF IPv6 working group mailing list
> ipv6@ietf.org
> Administrative Requests: https://www1.ietf.org/mailman/listinfo/ipv6
> --------------------------------------------------------------------

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