Hi Jeffrey,
A few responses below:
Zzh> Let me first describe two similar but not the same models - co-located AN
and UPF vs. integrated AN and UPF.
[John] co-located AN and UPF is what I referred to as "loosely coupled". I'm
fine with the co-located and integrated terminology as it is quite clear.
Zzh> In 5G, UPFs can already be very much distributed - as much as co-locating
UPF with gNB-CU. In this case, there is a local direct link between a gNB and
UPF or even just an internal logical link if gNB and UPF are virtual functions
running on the same compute server. With this, simplification of certain
functionality can already happen. For example, we don't need any special MEC
procedures anymore (UL traffic getting off a distributed UPF is no different
from getting off a central UPF - resources in local DN or central DN will be
applied). If the UPF implementation is based on routing/switching, then we
don't need special procedures for 5G-LAN either - existing IP/E-VPN solution
can be just used with wireless CE connections.
[John] I share the same view.
Zzh> Now, why do we bother further integrating AN and UPF together into a
single NF in xG?
Zzh> I realized that the draft does not talk about the following that I did
talk about in the presentation in last IETF:
(-- Optimized signaling and data plane / No need for separate N2 & N4
signaling / No need for AN-UPF connection)
[John] This is where it gets tricky.
For "optimized signaling" it implies a new interface that is composed of a
minimum of (N2 + N4) interface (new interaction model at SMF <----> ANUP, and
new interface)
This is a moving target because 3GPP model (and parameters) are being extended
/adapted/etc in every release.
For example, taking just QoS, Rel 18 work in XRM may end up adapting the QoS
flow model to further include sub-flow aspects. This in turn affects the
model/parameters at SMF, AN and UPF.
(and similar updates to model/parameters may happen across other work items in
3GPP)
My view is that this optimization cannot be done in IETF - its 3GPP work, and
even disregarding that it's a moving target (IETF will have to update this new
optimized interface on a 3GPP release schedule)
Zzh> By "loosely integrated" do you mean the "co-locating" model? If so, is
there any work needed at all?
[John] I think there may be some work like best practices for various
network aspects like optimized N3 connection, network domain security, perhaps
routing/steering/mobility.
For example, considering network sharing with multiple operators, optimal
interconnection while providing necessary isolation of the functions may be
useful to describe.
There may be other aspects as well that we can think about if this is of value
to explore further.
Would be happy to discuss sometime during the upcoming meeting too.
BR,
John
From: Jeffrey (Zhaohui) Zhang <[email protected]>
Sent: Monday, July 11, 2022 3:15 PM
To: Kaippallimalil John <[email protected]>; [email protected]
Subject: RE: [DMM] Questions/comments on draft-zzhang-dmm-mup-evolution-00
Hi John,
Thank you very much for the comments. Please see zzh> below.
Juniper Business Use Only
From: Kaippallimalil John
<[email protected]<mailto:[email protected]>>
Sent: Sunday, July 10, 2022 7:19 PM
To: [email protected]<mailto:[email protected]>; Jeffrey (Zhaohui) Zhang
<[email protected]<mailto:[email protected]>>
Subject: [DMM] Questions/comments on draft-zzhang-dmm-mup-evolution-00
[External Email. Be cautious of content]
Hi Jeffrey, all,
Thanks for putting together an interesting draft on mobile user plane evolution
and related IP network related aspects.
There are some important aspects that IETF/dmm could consider for xG without
directly impacting 3GPP architecture/standards.
This is partly described in comments under (3) below but would obviously need
much more discussion in the group.
Questions/comments for consideration:
1. Motivation for tightly integrating AN/UP functions in section 1.1 is not
fully clear
The point seems to be that with further virtualization and disaggregation, the
functions could be combined.
Sure, but the advantage of "combining" them is not clearly mentioned.
- for simplification (as for multicast, mobility, etc), perhaps the draft can
outline best practices to cut out unnecessary options, integrated management of
AN + UP.
- for performance, pseudo-wire as in 1.2 or some other dedicated high speed N3
connection should suffice (along with other recommendations in 3).
Zzh> Let me first describe two similar but not the same models - co-located AN
and UPF vs. integrated AN and UPF.
Zzh> In 5G, UPFs can already be very much distributed - as much as co-locating
UPF with gNB-CU. In this case, there is a local direct link between a gNB and
UPF or even just an internal logical link if gNB and UPF are virtual functions
running on the same compute server. With this, simplification of certain
functionality can already happen. For example, we don't need any special MEC
procedures anymore (UL traffic getting off a distributed UPF is no different
from getting off a central UPF - resources in local DN or central DN will be
applied). If the UPF implementation is based on routing/switching, then we
don't need special procedures for 5G-LAN either - existing IP/E-VPN solution
can be just used with wireless CE connections.
Zzh> Now, why do we bother further integrating AN and UPF together into a
single NF in xG?
Zzh> I realized that the draft does not talk about the following that I did
talk about in the presentation in last IETF:
Optimized signaling and data plane
* No need for separate N2 & N4 signaling
* No need for AN-UPF connection
Zzh> Instead of having AMF/SMF talk to gNB/UPF via N2/N4 separately, a single
control plane entity can talk the single integrated user plane entity. Getting
rid of AN-UPF connection is also good for latency besides reducing the
complexity.
Zzh> That's in addition to the benefit of a simplified architecture where
3GPP/wireless technology is for establish the radio link (just like an ethernet
cable or a WIFI connection connects a laptop to a router) and then leave the
rest to IETF/IEEE/wireline technology (then all IETF/IEEE/wireline solutions
can be applied w/o any work). Me being an engineer with IETF/wireline
background that may be biased but I think it makes sense and part of this
socializing in IETF is to see if it can get sympathy from the operators.
1. Concerns when combining AN, UP:
While the draft points out advantages of combining, there are some risks as
well. For example:
a) QoS - the basic QoS aspects that are conveyed via N4 to UPF (filters, rules,
etc) and other policies via NG-AP to RAN (and GTP-U /N3 in between) need to be
considered in depth.
Further, there are studies in 3GPP Rel 18 that are extending the QoS model,
what's handled on N4, policies to UPF or RAN and extensions to GTP-U.
Just saying that its all at ANUP does not address the issues (and
furthermore, this would be 3GPP work).
Zzh> In the co-locating model, the above needs to be addressed anyway; whatever
solution for that can be applied to the integration model, right? The only
difference is that the simple N3 (of co-locating model) is reduced to nothing
so no extensions to GTP-U is needed (actually it is not needed for co-locating
model either).
b) network sharing - in cases where, e.g., provider A operates AN and providers
B, C operates UPs, tight integration into ANUP can be inflexible.
(Reverse is also an issue, i.e., providers X, Y operate ANs and provider Z
operates UP)
zzh> Indeed - that's why the draft does have section 1.2 about having PW/GTP
between separate AN and UPF for three scenarios: a) MVNO b) Home-routed roaming
c) for whatever reason one may want to have a single UPF to handle several ANs
nearby.
Zzh> Having that option does not mean we don't need the integrated model.
Basically, we integrate when it makes sense and separate when it is necessary.
1. Other aspects
Consistent with the trend/direction to "softwarize" the different network
components, i.e., AN and UP in this case, IETF may be able to complement 3GPP
standards.
For example, scenarios where AN, UP functions deployed in an infrastructure
domain /data center and how security aspects, routing, QoS, mobility or
management of these group of functions are handled optimally.
(Note that in the above case, AN and UP are perhaps "loosely" integrated as
opposed to the "tight" integration in ANUP)
Zzh> By "loosely integrated" do you mean the "co-locating" model? If so, is
there any work needed at all?
Zzh> Thanks.
Zzh> Jeffrey
BR,
John
_______________________________________________
dmm mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/dmm
