Re: [tsvwg] 5G, DiffServ and new PHBs

[Dave Taht <[e-mail]>]

On Tue, May 12, 2020 at 6:06 AM DOLLY, MARTIN C <[e-mail]> wrote:
>
> QCI’s with lower latency will be emitted first by the eNB

One of the things that seem to be underdefined (and left to the
operator) are the number of retries allowed for reliable transport at
this layer. I think for the voice queue it's in the range 8-100ms? For
the BE queue?

Are there any "typical" parameters for this?

Are there measured results for the typical retry distributions in
various circumstances?

>
>
>
> From: tsvwg <[e-mail]> On Behalf Of Jerome Henry (jerhenry)
> Sent: Monday, May 11, 2020 7:07 PM
> To: [e-mail]; [e-mail]
> Cc: [e-mail]
> Subject: Re: [tsvwg] 5G, DiffServ and new PHBs
>
>
>
> Thank you Ruediger.
>
>
>
> I am not sure to fully understand why using a QCI would remove spectrum (it would be congruent to say that DSCPs remove bandwidth on the Internet). I have been thinking about it for a good part of the day, and think that your colleague may be referring to the idea of your second paragraph, that if we allocate more QBRs, then we mechanically reduce access to other traffic, in order to ensure a guaranteed bearer. And if we guarantee bandwidth, we need to make sure that the BW is provided, which also mechanically constraints the access of other traffic over the same radio.
>
>
>
> This definitely sounds consistent with what I hear as well. I must say that I have also heard a couple of times the statement “we implement Voice and BE, and don’t care about the rest” from some implementers. But I have also heard very different positions from many others. It seems that the disconnect may come from the notion of small number of users vs bigger number. In essence, it seems to me that a large cell deployed to connect a gazillion of UEs across a very large area will attempt to maximize the overall aggregated throughout. In this environment, I expect BE to be the norm, and Voice QCI to be a necessary luxury. As the cell size decreases, as deployment density (from one or more implementers in the same physical area), the conversation tends to shift to SLA as part of the value proposition. In these environments, I hear the “2 QCI” argument much less often. Obviously, I will not provide specifics, but I will say that implementations do exist, and they are not uncommon in my universe (thus the need for the mapping, it is not an academic exercise).
>
>
>
> I remember, after hearing this point (“really only 2 QCIs in the field”) and asking my colleagues working in 3GPP how general this statement was (“it depends” and “it is far from being a universal truth” was their answer), I went to ask more generally (non-Cisco) 3GPP folks if 3GPP had been working, since release 8, year after year, release after release, to add new QCIs consistently, redesigning their scheme deeply with 5G, to put together a growing list of QCIs no one in fact really used or cared about. I think I still have planted in my back a butter knife that was thrown at me as I was running away. I am half joking, but you see my point. If there was 0 adoption, the effort to define more QCIs and deeply rethink the structure for 5QI would likely not have happened. So when I hear that adoption is there, but depends on the location and the implementer, I hear something that is consistent with what I hear about DSCP.
>
>
>
> My 2 cents
>
>
>
> Jerome
>
>
>
>
>
>
>
> From: "[e-mail]" <[e-mail]>
> Date: Monday, May 11, 2020 at 4:31 AM
> To: "Jerome Henry (jerhenry)" <[e-mail]>, "[e-mail]" <[e-mail]>
> Cc: "[e-mail]" <[e-mail]>
> Subject: AW: 5G, DiffServ and new PHBs
>
>
>
> Hi Uma, hi Jerome,
>
>
>
> after having had a brief discussion with a colleague working on mobile service development, I’d like to add a piece of information. He mentioned that deployment of QCIs comes at a cost of coding space / wireless spectrum. To better utilize wireless spectrum, so far deployment is often limited to one QCI suitable for telephony and one for Best Effort. He doesn’t expect that to change.
>
>
>
> He mentioned that supporting more QCIs gives those in some cases guaranteed bandwidth, which will cause blocking of other transmissions until the QCI served terminal received its bandwidth (my terminology might be less than optimal here). The resulting effect is, that a small number of users experience fair Quality of Service, while a bigger number users faces a service degradation.
>
>
>
> I recommend to verify that. If that holds, I’d appreciate appropriate text in IETF work relating DiffServ and 3GPP QCIs. To me one point to be discussed then is, whether IETF should set aside reserves or build mechanisms to support technologies of other SDOs, also if the latter are not deployed at significant scale.
>
>
>
> Regards,
>
>
>
> Ruediger
>
>
>
>
>
>
>
>



-- 
"For a successful technology, reality must take precedence over public
relations, for Mother Nature cannot be fooled" - Richard Feynman

[e-mail] <Dave Täht> CTO, TekLibre, LLC Tel: [phone]