Re: [tsvwg] Some comments on NQB (part 2)

[Greg White <[e-mail]>]

Hi David,

Much of the answer to your question can be found in Section 8.3 of the draft (https://www.ietf.org/archive/id/draft-ietf-tsvwg-nqb-10.html#section-8.3) along with my post to the TSVWG mailing list in February (https://mailarchive.ietf.org/arch/msg/tsvwg/fBcm6eUfv9egmlo-TX8N3OAFM_A/).  Please have a look at those.

In response to your blunt statement, I think that the vast majority of Internet users and application developers would be confused by it, and think that you have it exactly backward.  For them, Internet QoS is pure fiction.  From their perspective, there has never been such a thing as Internet QoS and it is unlikely that there ever will be.  I've tried to describe my view as to the reasons for this in Section 3.2 of the draft (https://www.ietf.org/archive/id/draft-ietf-tsvwg-nqb-10.html#section-3.2).   On the other hand, there are hundreds of millions of Wi-Fi networks around the world that implement WMM using the default DSCP-to-UP mapping.  The QoS treatment in these networks is real, and these networks carry a tremendous amount of traffic for applications used every day by billions of people. Moreover, my experience (and I think this view is shared by others), is that the most common sources of quality degradation (loss, latency, jitter) for Internet users are at the edge (Wi-Fi and access) rather than the core. So, it's imperative that the solution solves for Wi-Fi. The characterization of these networks as "legacy WiFi" makes it sound as if we're talking about older gear that will soon be replaced by gear that would allow the use of some other DSCP (e.g. 5) to achieve the desired outcome. This is not true.  Most brand new Wi-Fi 6 gear available for purchase today implements the same default mapping.  For the future, the WFA has recently begun to recommend that Wi-Fi devices use the mapping in RFC8325, but it is unclear how quickly that will get adopted, implemented, and deployed, and furthermore it is unclear whether the WFA means RFC8325 "as published" or RFC8325 "with updates" (RFC8622 already updates it, and NQB is proposed to).    Regardless of which interpretation manufacturers take, it doesn't create a preference for the value 5. 

This "issue" of re-marking DSCPs is to me a necessary component of trying to (today) enable a truly end-to-end PHB across the Internet. The majority of Internet networks, operating systems, and applications that I'm familiar with do not implement the IETF standardized PHBs and Service Classes, nor do they follow the recommended DSCP assignments from the IANA registry. The likelihood of finding an end-to-end path that has consistent DSCP interpretation is probably fairly close to zero in my view (even the "default" DSCP of 0 doesn't mean default treatment in some major ISP networks!).  And, as Jason indicated, the majority of traffic to access networks flows over direct interconnect links anyway, where DSCP usage can be coordinated and managed.

So, for me the critical part of the DSCP assignment for NQB is to provide guidance to application and OS developers that will enable interoperation with the established (and largely unmanaged) practices at the edge.  What happens in the core will almost certainly involve some level of DSCP re-marking and communication between interconnecting networks, regardless of which DSCP we pick.  If we can provide recommendations in the draft that make it easier for interconnecting networks (and perhaps limit the amount of re-marking needed), then we absolutely should do it (and for that some seem to prefer 45 because it aligns with usage at the edge, others 5 because they prefer IP precedence), but we can't let *that* "tail" wag the dog.
 
-Greg
 

On 5/4/22, 2:08 PM, "Black, David" <[e-mail]> wrote:

    <WG_Chair_Hat=OFF>
    With apologies for resurrecting some topics that have been more or less settled in the past, I am still bothered by the recommendation of two default DSCPs for NQB.

    The question that I can't satisfactorily answer is: If NQB traffic is supposed to be carried as a peer to Default traffic, why are we instructing end systems to use DSCP 45 for originated NQB traffic on all networks?

    The answers to that question seem to boil down to (with apologies for the crass bluntness) necessity of allowing the Legacy WiFi "tail" to wag the Internet QoS "dog".

    Would someone (Greg?) provide a reminder of what is it about legacy WiFi that requires this approach, please ...
    </WG_Chair_Hat>

    Thanks, --David

    -----Original Message-----
    From: tsvwg <[e-mail]> On Behalf Of Greg White
    Sent: Friday, April 29, 2022 3:34 PM
    To: [e-mail]
    Cc: [e-mail]
    Subject: Re: [tsvwg] Some comments on NQB (part 2)


    [EXTERNAL EMAIL] 

    Thanks Ruediger.

    Glad to hear that we are converging, though it wasn't clear to me which version of the new text you preferred.  For now, I'll stick with the version that I'd sent on April 4, but let me know if I've misunderstood you.
    Hopefully others find this text change acceptable.  

    N.B. I don't have any issue with your bigger picture idea, but it is beyond the scope of the NQB draft.   So, if you want to pursue documenting it in an RFC, it probably should be proposed separately.  

    So, for the NQB draft, are folks ok with replacing:

    To support NQB, networks MUST preserve a DSCP marking distinction between NQB traffic and Default traffic when forwarding via an interconnect from or to another network. 
    To facilitate the default treatment of NQB traffic in backbones and core networks discussed in the previous section (where IP Precedence may be deployed), networks that support NQB SHOULD NOT use the value 45 for NQB at network interconnects unless that usage is explicitly documented in the TCA (Traffic Conditioning Agreement, see [RFC2475]) for that interconnection. 
    Rather, networks SHOULD remap NQB traffic to DSCP 5 prior to interconnection, unless agreed otherwise between the interconnecting partners. 
    To be clear, interconnecting networks are not precluded from negotiating (via an SLA, TCA, or some other agreement) a different DSCP to use to signal NQB across an interconnect. 
    Additionally, the fact that this PHB is intended for end-to-end usage does not preclude networks from mapping the NQB DSCP to a value other than 45 or 5 for internal usage, as long as the appropriate NQB DSCP is restored when forwarding to another network.


    With [notes in square brackets added to help those trying to compare against the above]:

    To support NQB, networks MUST preserve a DSCP marking distinction between NQB traffic and Default traffic when forwarding via an interconnect from or to another network.    [no change]
    Networks that support NQB SHOULD support the ability to re-mark NQB traffic prior to such an interconnection.    [new recommendation]
    It is RECOMMENDED that interconnecting networks negotiate the use of the DSCP value 45 to indicate NQB traffic across their interconnections (thus avoiding the need to re-mark traffic), however, local DSCP usage by either network could require the use of a different value.   [new recommendation]
    To be clear, interconnecting networks are not precluded from negotiating (via an SLA, TCA, or some other agreement) a different DSCP than 45 to use to mark NQB traffic across an interconnect.  [only editorial change]
    In situations where negotiation of a DSCP between interconnection partners is infeasible, networks that support NQB SHOULD NOT use the value 45 for NQB at network interconnects, but rather SHOULD re-mark NQB traffic to DSCP 5 prior to interconnection.  [limited the applicability of this recommendation]
    This is intended to facilitate the default treatment of NQB traffic in backbones and core networks discussed in the previous section (where it is possible that IP Precedence may still be deployed).  [only editorial change]
    Additionally, the fact that this PHB is intended for end-to-end usage does not preclude networks from mapping the NQB DSCP to a value other than 45 or 5 for internal usage, as long as the appropriate NQB DSCP is restored when forwarding to another network.  [no change]


    In addition to Ruediger, I'd like to specifically hear from David Black and Gorry, since two of the original sentences came from David, and Gorry was the OP raising a concern about those sentences.


    -Greg



    On 4/29/22, 3:27 AM, "[e-mail]" <[e-mail]> wrote:

        Hi Greg,

        Thanks. My bigger picture: RFC 8100 is intended to support interconnection policies like:

        If DSCP in range <0-7>      # with a negotiated SLA, different ranges may apply for different backbone PHBs and Codepoint rewrites...
            then PHB=default

        To me, no SLA negotiation is necessary if forwarding expected by a backbone is "default". In addition, I prefer the interconnection QoS policy to be as simple as possible, if no QoS SLA is negotiated:

        (PHB=default)
        If DSCP in range <8-63> 
            set DSCP=000 000

        I appreciate your suggested text which allows that; no DSCP 45 traffic should be received at interconnections without negotiated QoS SLA, if the above is deployed. If DiffServ Standards were changed to support the above, 8 DSCP are available for PHBs whose differentiating support is most useful and can be decided upon at the access. I think that would be beneficial for
        - Lower Effort PHB
        - L4S / NQB (I think, any DSCP can be rewritten at access and may be at a home gateway, and if a standard proposes a value like 45, the better).

        What I suggest to avoid at interconnection (and will not deploy, where I'm in charge) is (e.g.):

        If <InterconnectionPartnerX> then
            If DSCP <a> 
                then PHB=default
            If DSCP <b> 
                then PHB=default AND set DSCP=000 001
            If DSCP <c> 
                then PHB=EF
            If DSCP <d> 
                then PHB=AF4 AND set DSCP=001 010
            If DSCP <e> 
                then PHB=AF4 AND set DSCP=001 100
            If DSCP <f> 
                then PHB=default AND set DSCP=000 101
              ....
            If <no match>
                then PHB=default AND set DSCP=000 000

        Rather than individual per interconnection partner combined with per DSCP policies at interconnection, I'm looking for simplistic, easily comprehensible and to the extent possible generic Behaviour Aggregate classification. That holds for (range based DSCP) remarking at interconnection too.

        Regards,

        Ruediger





        -----Ursprüngliche Nachricht-----
        Von: Greg White <[e-mail]> 
        Gesendet: Freitag, 29. April 2022 01:12
        An: Geib, Rüdiger <[e-mail]>
        Cc: [e-mail]
        Betreff: Re: [tsvwg] Some comments on NQB (part 2)

        Hi Ruediger,  

        Thanks for responding.   See my responses [GW] below.

        -Greg


        On 4/26/22, 8:14 AM, "[e-mail]" <[e-mail]> wrote:

            Greg,

            Sorry for the late response.

            <snip>

            You wrote:
            What I offered for consideration below was that the [DSCP] value 45 be recommended across interconnections in cases where the two interconnecting partners are NQB-aware and they negotiate DSCP markings.

            RG: To me, that's an additional concept. My take was, NQB doesn't require more than default transport in the backbone and at interconnection. If the later holds, negotiation of NQB is no issue to me, but an appropriately picked DSCP is important (it should unambiguously indicate "default forwarding" at interconnection).
            If a QoS SLA is negotiated, in principle any negotiated DSCP does (it is well known that I prefer RFC 8100 at wholesale and interconnection interfaces, as this simplifies deployment and operation).

        [GW] It seems to me that there isn't such a thing as a DSCP (other than possibly 0) that unambiguously indicates default forwarding at interconnection.  I quickly re-read RFC8100 and also don't see mention of it there (it refers to DSCP=0 as being default and seems to recommend that any traffic classified into the Default / Elastic Treatment Aggregate be re-marked to 0). As I understand it, the practice of aggregating traffic based on the IPP bits (top 3 bits) is not universal. If I'm right in that, then it seems that recommending NQB-aware networks re-mark NQB traffic to 5 and not use 45 at *all* interconnections might be unnecessary (and it was apparently concerning to some).  
        In my post on April 4: https://urldefense.com/v3/__https://mailarchive.ietf.org/arch/msg/tsvwg/PKTrfNdTCEXmwoovSkqec6cOJZc/__;!!LpKI!mzr4n4KDAty5Aq0a1tG2B89wGfRac3ylHv0FS_U75V-j47XXXS_4VgGjl_ncHFL_4IO4sSMU0X0akSepO1Y$ [mailarchive[.]ietf[.]org] in response to Gorry's concerns, I had suggested softening this to (paraphrasing here):
        - If negotiating a DSCP to use at interconnection, recommend 45, but the parties can negotiate whichever value they want.
        - If negotiation isn't possible, the sending network SHOULD NOT use 45, and instead SHOULD use 5. 
        What about this do you not like?  It seems to me that you're saying that you wouldn't negotiate a DSCP for NQB.  So, based on the proposed text, your interconnection partners SHOULD use 5. 
        Would it make you happier if the first statement were replaced with:
        - If negotiating a DSCP to use at interconnection, recommend the use of either 5 or 45, but the parties can negotiate whichever value they want.



            You wrote:
            The data from Ana Custura and Gorry indicates that, unless something changes in regards to bleaching of the upper 3 bits by some networks, any future assignments of the values 13, 21, 29, 37, 53, 61 would do well to keep in mind that any traffic so marked could end up being aggregated with NQB traffic.  That said, this sort of bleaching is non-compliant with the definition of the DSCP field, and is already problematic for EF, VA, and all of the CS codepoints (which aggregate in incompatible ways), so (as was commented in the last meeting) we may want to consider identifying the routers that continue to do this, and try to work with the associated network operators to change the behavior. 

            RG: I'd appreciate a concise reference for your claim "this sort of bleaching is non-compliant with the definition of the DSCP field".

        [GW] I probably didn't choose my words as carefully as I could have, and I made that statement (without doing the appropriate research) based on comments others had made.  RFC2474 Section 3 seems to imply to me that selectively bleaching certain bits of the field is not what was intended, but it does allow that "Nodes MAY rewrite the DS field as needed to provide a desired local or end-to-end service."  So, I don’t see any requirement statement that is violated. 


            RG: If you are interested, I can sketch examples where single sided changes were made to well  negotiated EF deployments and the interesting consequences caused by that. That's not what mean by "problematic for EF", it rather shows what happens if a QoS design isn't well agreed with all parties responsible for QoS aware network sections and policy points in an operational end-to-end production chain.

            Regards,

            Ruediger