Re: [tsvwg] residential broadband BCP PHB and CP treatment Re: CC/bleaching thoughts for draft-ietf-tsvwg-le-phb-04

[Brian E Carpenter <[e-mail]>]

On 21/04/2018 17:29, Mikael Abrahamsson wrote:
> On Sat, 21 Apr 2018, Brian E Carpenter wrote:
> 
>> Because RFC2474 clearly states that CS1 gets precedence over DF (=CS0).
>> (Which in fact goes back to RFC791 and RFC1349.)
>>
>> "PHBs selected by a Class Selector Codepoint
>> SHOULD give packets a probability of timely forwarding that is not
>> lower than that given to packets marked with a Class Selector
>> codepoint of lower relative order..." (RFC2474 section 4.2.2.2)
>>
>> None of the other RFCs that use the CS1 codepoint for LE have formally
>> updated RFC2474; in other words they all violate the SHOULD above.
>> Specifically, RFC4594 did not formally update RFC2474.
>>
>> A user who sets CS1 because they want better treatment than default
>> is doing exactly what an IETF standard says. So if you see CS1 on
>> incoming traffic, you have no idea whether the user wanted Precedence 1
>> or LE, unless you have a specific understanding with that user.
>>
>> To me that means that the correct default behaviour is to map CS1 to
>> DF, as a compromise. Of course, if you *know* whether the user is
>> relying on RFC2474 or on RFC4594, you can configure the correct
>> mapping. But that's always true of diffserv.
> 
> Thanks for the explanation. Would it influence your opinion if there were 
> data that showed that more applications use CS1 in the LE sense compared 
> to using CS1 as higher preference over DF?

It doesn't change the basic problem that we have actually published
inconsistent standards, sadly. I really feel bad that I didn't make a fuss
about this earlier, but it was a boiling frog situation, starting with
RFC3662 being only Informational and only "suggesting" a codepoint.

> Basically my question is that if setting formal references/updates etc 
> aside, what have people actually been doing in the past 20 years, should 
> that affect our decision?

It should certainly affect operational recommendations. So I think you're
correct that a BCP or nearly-BCP is needed, explaining the issues and
defining (a) default configuration and (b) the necessary configuration
options. I also think we need to be sure that we don't create yet more
ambiguity about the CS1 codepoint. The le-phb-04 text removes the treatment
of CS1 from RFC4594, but gives no guidance about CS1 as defined by RFC2474.
At least we could say that it's operationally ambiguous and should not
be used as a default setting in any host.

> Discovering that for instance ssh uses IPTOS by default was a revelation 
> to me.

Yes. Of course, it was partly for compatibility with legacy code that
RFC2474 specified that all DSCPs were mutable at domain boundaries.

    Brian