Re: [tsvwg] Delay vs. Loss draft: Terminology

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

On 18/07/2013 09:37, Black, David wrote:
> <WG chair hat off>
> 
> This note concerns draft-polk-tsvwg-delay-vs-loss-ds-service-classes
> 
> I believe I see some sloppy use of terminology in this draft and the
> related discussion.
> 
> CS4 was originally defined as a DSCP value in RFC 2474 (Diffserv field
> in IP headers) and is used only as a DSCP value in RFC 4594 (Diffserv
> Service classes).
> 
> That said, RFC 2474 can be read to allow use of CS4 as a PHB name - it's
> effectively shorthand for "the PHB assigned to the CS4 DSCP - as is done
> in draft-polk-tsvwg-rfc4594-update does.  While that may not be the best
> usage, I think it's ok.
> 
> OTOH, draft-polk-tsvwg-delay-vs-loss-ds-service-classes appears to be
> using strings such as "CS4", "CS4-Discardable" and "CS4+" as service
> class names (see Sections 5 and 6).  That inclusion of the DSCP name
> in the service class name is (IMHO), a stunningly bad idea that is
> going to lead to confusion, and moreover is at odds with one of the basic
> principles of the Diffserv architecture, namely separation of traffic
> behavior (PHB and subsequently, service class) from the DSCP used in
> the IP header to request suitable treatment in a particular network.

Hear, hear! It's especially true because CSn was included explicitly
for backwards-compatibility with RFC 791, so attempting to name new
behaviours as derivatives of CSn is particularly misleading. If you're
looking for a PHB that's like CS4 but "more discardable" that would be
CS3, anyway.

The CSn, including CS0, are unique in that the assigned code
points are not *recommended* mappings - section 4.2.2.1 of
RFC 2474 reserves them unconditionally. No wiggle room here;
there's a 1:1 correspondence between DSCP CSn and PHB CSn.
But that applies nowhere else in diffserv.

    Brian

> As an author of both RFC 2474 and 2475, I really would like to
> see draft-polk-tsvwg-delay-vs-loss-ds-service-classes revised to be
> clear on DSCP vs. PHB vs. service class and in particular to cease
> and desist from using CS4 (or any other DSCP name) as the name or part
> of the name of a service class - that will inevitably lead to the bad
> and wrong assumption that such a service class can only ever be
> deployed with the CS4 DSCP - while that sort of relationship is
> RECOMMENDED by RFC 4594, it is not REQUIRED (where "RECOMMENNDED"
> and "REQUIRED" have their RFC 2119 meanings).
> 
> An example of the possible confusion that is that there will be
> networks in which the CS4 DSCP will be deployed with only one
> of the service classes currently called "CS4" and "CS4-Discardable" in
> draft-polk-tsvwg-delay-vs-loss-ds-service-classes.  Is anyone prepared
> to try to explain with a straight face why the "CS4" service class
> isn't using the "CS4" DSCP?
> 
> Thanks,
> --David
> ----------------------------------------------------
> David L. Black, Distinguished Engineer
> EMC Corporation, 176 South St., Hopkinton, MA  01748
> [phone]             FAX: [phone]
> [e-mail]        Mobile: [phone]
> ----------------------------------------------------
> 
> 
>