John,

In line comments...

On 2011-03-10 04:48, John Leslie wrote:
> Brian E Carpenter <[email protected]> wrote:
>> I'm not hearing much in response to the open issue:
>>
>>   Should we delete most of the text concerning stateful methods
>>   of handling the flow label?
> 
>    I don't really have a horse in that race, but at first blush I'd say
> the I-D goes into too much detail about stateful handling.
> 
>    IMHO, most handling (assuming it happens at all) will be stateless,
> and will greatly resemble "damage" (as in what we route-around).
> 
>> In fact I'm not hearing much at all. Should I ask the WG Chairs
>> to conclude that everybody is OK with these drafts? That would
>> save us some time in Prague...
> 
>    I won't be in Prague... :^(
> 
>    But while this I-D is a step in the right direction, where I think
> we ought to get should allow much more leeway in setting -- even
> changing -- Flow Labels _by_prior_arrangement_.

The draft creates that leeway - apart from the Section 4 text about
stateful methods.
> 
>    The basic assumption that Flow Labels will be set by the sender and
> passed unchanged is certainly correct in the absence of arrangements
> that specify particular processing.
> 
>    I am a bit nervous about _always_ allowing a zero Flow Label to be
> changed without sender consent...
> 
>    But clearly, there _can_ be no protection against such changes --
> or any changes, for that matter.
> 
>    My intuition is that any time the Traffic Class is changed during
> forwarding, rules about Flow Label particular to that Traffic Class
> should apply.

There was considerable pushback in the WG to adding the concept
of local domains to the flow label definition.

> 
>    Further, there probably need to be _various_ largely out-of-band
> mechanisms to validate a contractual basis for special handling of
> packets based upon Flow Labels. There could be one or more in-band
> mechanisms if we had what I'll call a "Middlebox Options" extension
> header (between Hop-by-Hop and Destination Options) to store validation
> information.

Yes, but that's in the stateful part that you're arguing above
should be replaced by "more leeway".

> 
>    Fundamentally, I don't believe the completely-insecure Flow Label
> can serve by itself as sufficient justification for special handling;
> and I don't believe that the sender alone can justify special handling
> by _all_ the different service-providers along the forwarding path.
> I think more infrastructure will need to evolve.

Ditto.

> 
>    Legislating what we can't specify any means to enforce isn't wise...
> 
> ====
> 
>    Disclaimer: I in some sense "represent" the ConEx WG, where several
> folks want to reserve one bit from the Flow Label for ConEx use. I am
> _not_ the right person to put the case for that, because frankly I
> don't agree with the idea. Nonetheless, it seems reasonable to me to
> allow things like that on an Experimental basis if clearly labeled
> by setting an Experimental Traffic Class.

Ditto. What the draft is talking about is default behaviour, absent
anything requiring prior agreement or signalling.

I take your whole comment as a recommendation to remove section 4.

     Brian
--------------------------------------------------------------------
IETF IPv6 working group mailing list
[email protected]
Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6
--------------------------------------------------------------------

Reply via email to