Re: Tunnels and diffserv question

[Brian E Carpenter <brc@zurich.ibm.com>]

Rémi Denis-Courmont wrote:
> Le samedi 26 août 2006 03:14, vous avez écrit :
>
> > The abstract mentions "IP tunnels of various forms", section 6
> > explicitly mentions "IPv6 over  IPv4 tunnels (RFC1933)". This in
> > special and the rest of the text makes it seems to be fully
> > applicable to all the other transition mechanisms. Simply copying
> > over the inner ToS/DS field to the outer one should be sufficient.
>
>
> Right, thanks. That's the first conceptual model from §3.1. I was merely 
> wondering if it was recommended that 6to4, Teredo (and ISATAP?) use the 
> first or the second model... it does not make much sense, I think, to 
> have different model implemented by different nodes since they are all 
> part of the same multi-point pseudo-tunnel.
>
>
> > I haven't seen any implementation actually doing this though, but it
> > could be that some implementations already do it in that manner.
>
>
> Anyway, I'm afraid this is hardly implementable in practice, since there 
> is no IPv4-equivalent of the sendmsg() IPV6_TCLASS option, but that's 
> slightly off-topic.

That shouldn't matter, assuming the tunnel end-point is below the socket
API or somewhere inside the network. It should map the DSCP regardless of
who set it. Today, DSCPs are usually set by policy or by a classifier,
not by an API.

    Brian