5G Distributed UPFs for 5G Multicast and Broadcast Services (5MBS)

Distributed UPFs in 5G User Plane Mobile User Plane (MUP) in 5G has two distinct parts: the Access Network part between UE and gNB, and the Core Network part between gNB and UPF. UPFs are traditionally deployed at central locations, with UEs' PDU sessions encapsulated and extended thru GTP-U tunnels via the N3 (and potentially N9) interfaces in 5GS. The interface N6 supports fundamentally a direct IP or Ethernet connection to the data network or DNN. Actually, UPFs could be distributed & deployed closer to gNBs.
The draft has described the 5G mobile user plane (MUP) via the refinement of distributed UPFs or dUPFs. The following picture shows the dUPF architecture: In distributed UPF architecture, the central (PSA) UPF is no longer needed. dUPF1 and UPF2 connect via PE1 and PE2, respectively, to the DN VPN (or network instance/NI) that UE1 and UE2 intend to access. There could exist other PEs, like PE3 in the picture, for other sites of the same network domain(VPN or NI) or for global Internet access. There are some benefits of distributed UPFs:

The N3 interface becomes very simple - over a direct or short transport connection between gNB and dUPF.
The transport infrastructure off N3/N9 and N6 are straightforward, most likely over the same underlay VPN (MPLS, SR-MPLS or SRv6) supporting the traditional N3/N9 tunneling as in centralized PSA UPF case.
MEC becomes much simpler since no need to deploy centralized PSA UPF plus ULCL UPFs; UE-UE traffic can be optimized for LAN-type services (via host-route).

In short, the distributed UPFs model achieves "N3/N9/N6 shortcut and central UPF bypass", which is desired by many operators.

5G Multicast and Broadcast Services (5MBS) The 3GPP document TS 23.247 for 5G multicast and broadcast services, or 5MBS, specifies the 5GS architecture to support MBS communication. The following picture shows the brief system architecture of 5MBS: TS 23.247 adds new 5GS network functions (NFs) on both 5G control-plane (CP) and user-plane (UP). For example, the CP NF MB-SMF is, in collaboration with the regular SMF, to provision and signal to the UP NF MB-UPF (via the interface N4mb) for setting up MBS delivery path. 5MBS has specified two data delivery modes, individual delivery vs. shared delivery:

Individual delivery: When the (downlink or DL) MBS packets are received by the MB-UPF from the interface N6mb, MB-UPF replicates & forwards those packets towards (multiple) UPFs, via the interface N19mb, through either unicast (requiring multiple GTP tunnels if unicast underlay transport is applied) or multicast (if multicast underlay transport over N19mb is applied) transmission.
Shared delivery: When the (DL) MBS packets are received by the MB-UPF from N6mb, MB-UPF replicates & forwards those packets towards (multiple) gNBs, via the interface N3mb (the lower-path in the picture), through either (multiple) separate GTP tunnels if unicast underlay transport over N3mb is applied, or a single GTP tunnel if multicast underlay over N3mb is supported.

Challenges in 5G MBS Communication

5MBS Transport Challenges The 5MBS architecture in TS 23.247 introduces some network challenges:

Because of the addition of new CP and UP NFs, this will post additional provisioning & implementation challenges to the underlay transport infrastructure. For example, in the individual delivery mode, both SMF and MB-SMF have to synchronize with each other to help set up the relay/stitching path between UPF, MB-UPF and DN.
The picture in previous section shows three new interface types corresponding to three different segments: N3mb, N6mb and N19mb. Based on the traffic delivery mode, once MB-UPF receives DL traffic from N6mb, it will have to do either individual or shared delivery.
In accordance with TS 23.247 , the underlay transport infrastructure of all three segments can use either unicast or multicast transmission, based on the capabilities of underlay networks. For example, for the DL shared delivery from MB-UPF to gNB via the interface N3mb, 5G MBS packets can be transmitted to multiple gNBs via multicast transmission if the underlay network supports. Otherwise, MB-UPF will have to use unicast to transmit separately to (multiple) gNBs. Considering that this unicast/multicast flexibility is applicable to all the three above-mentioned segments, the implementation will have to face more challenges.

5MBS UP Signaling Challenges The user plane from the MB-UPF to gNB directly (i.e., the lower-path in the above figure for the shared delivery) and the user plane from the MB-UPF to UPFs then to gNB (i.e., the upper path in the figure for individual delivery) may use IP multicast transport via a common GTP-U tunnel per MBS session, or use unicast transport via separate GTP-U tunnels at gNB or at UPF per MBS session. When using the IP multicast transport, GTP-U Multicast Tunnels shall be used for unidirectional transfer of the encapsulated T-PDUs from one GTP-U Tunnel Endpoint (i.e., acting as the sender) to multiple GTP-U Tunnel Endpoints (i.e., acting as receivers). The Common Tunnel Endpoint ID (C-TEID) which is present in the GTP header shall indicate which tunnel a particular T-PDU belongs to. The C-TEID value to be used in the TEID field shall be allocated at the source Tunnel Endpoint (e.g., the sender) and signaled to the destination Tunnel Endpoints (e.g., receivers) using a control plane protocol, e.g., GTPv1-C & GTPv2-C. One C-TEID shall be allocated per MBMS bearer service or per MBS session . As we have explained in the draft , the signaling overhead to establish a N3 GTP unicast tunnel has reached seven steps, let alone the case of the more complicated MBS tunnel creation.

5MBS Challenges in Satellite Communication The 5G service via the satellite constellation has become a popular topic in 3GPP. There are currently three major satellite-related projects in SA workgroups, i.e., the satellite access (SAT_Ph2) and the satellite backhaul (SATB) in SA2 as well as the Phase-3 enhancement via the satellite-based store-and-forward technology (SAT_Ph3) in SA1 WG . These projects study various 5GS requirements when either a gNB or a UPF or both are on-board satellites. Evidently, the continuously-moving satellite constellations introduce another dimension of challenges to UE registration, session management and traffic routing. The GTP-U tunnel end points have to be changed frequently when the satellite providing the on-board service for a UPF rotates away from the corresponding gNB of the same GTP-U tunnel. For the SAT_access case, the ground station (GS) has to find a new gNB on-board another satellite every couple of minutes (e.g., being around 7-8 minutes for the LEO category) to hand over UEs. There are significantly large amount of singalling messages involved even for unicast case via satellite constellation, let alone if we extend the similar scenarios to 5G MBS communication.

5G Distributed UPF for 5G MBS Implementation The REQ8 of talks about the multicast efficiency between non-optimal and optimal routes, where it states that, in term of multicast considerations, DMM SHOULD enable multicast solutions to be developed to avoid network inefficiency in multicast traffic delivery. The current 5MBS architecture requires all DL multicast traffic go through the (centralized) MB-UPF, regardless of using the individual or shared delivery. In many operators' networks, 5GS might be deployed in a location that is distant from customer sites. If the deployed site happens to be on-board satellites, the additional complexities and moving dynamics will certainly worsen the operations. In these scenarios, the efficiency of multicast transmission will be compromised. On the other aspect, a 5G dUPF deployed closer to gNB, or even more radically applying 'ANUP' via the possible integration of gNB & UPF , might lead to more efficient implementation:

For shared delivery, the MB-UPF can be distributed closer to or integrated with gNB, i.e., either dUPF or ANUP-like. The N6mb is a normal IP interface which is connected to DN over underlay network. This transport connection will most likely use the VPN infrastructure that has been provisioned by operators for 5GS. As a dUPF or ANUP, the N3mb tunnel off MB-UPF could be made much simpler. In some field edge sites, a dUPF could co-locate on-prem with gNB, which can even remove the usage of complex (inter-site) VPN to favor native IP transport.
For individual delivery, it involves two UPFs, one regular UPF and one MB-UPF. To follow the current 3GPP specification, we can distribute and deploy both UPFs closer to gNB. While the DL traffic off the N6mb interface may achieve the same gain as in the shared-delivery mode, the transport for the N19mb tunnel and the (regular) N3 tunnel can be significantly simplified. Remember we have mentioned previously that either unicast or multicast (underlay) transmission can be used for N19mb (and actually also for N6mb and N3mb). Therefore, applying dUPF or, possibly ANUP in future, will help simplify the N19mb VPN transmission.
For individual delivery, if we expand the scope beyond the current 3GPP spec., e.g., looking beyond the 5G or even 6G roadmap that are already on the horizon of the 3GPP planning, we could integrate the regular UPF and MB-UPF together as a distributed UPF, and then deploy the dUPF closer to gNB. Of course, we might even take one step further by integrating both UPFs (UPF and MB-UPF) and gNB as a single 'logical' node, i.e., ANUP . Regardless, in either scenario, both the N19mb and N3 tunnels can be simplified, or even consolidated, significantly, TS 23.247 specifies the behaviors of MB-UPF, as a standalone NF. Indeed, all the features and behaviors that would be implemented by a MB-UPF can be collaboratively integrated into a regular UPF. This type of 'merging' should lead to more network efficiency and better multicast traffic forwarding, conforming to the REQ8.

When we take into consideration the above plausible arguments and accordingly apply them to different 3GPP satellite-related projects, e.g., SATB (backhaul), SAT_Ph2 & SAT_Ph3 (access), we can certainly draw the conclusion that the extra burden of signalling messages, the complexity of control plane as well as the excessive encapsulations of user plane, as introduced by 5MBS, can be relieved dramatically. Both drafts discussed and compared briefly different tunneling mechanisms to implement the 3GPP GTP-U UP, i.e., SRv6, MPLS as the underlay, or in specifying a new SRv6 based MUP architecture to replace the GTP-U. While these proposals may experience different issues upon 5MBS transport implementation, the application of distributed or 'integrated' UPF might make it more feasible.

Security Considerations TBD.

IANA Considerations This document requests no IANA actions.