The 6G Network is Flat

I am delighted to be able to write to you, after having filed a new patent application that should bring us true Ultra Low Latency Communications in between 4G, 5G and 6G user equipment. Some future iPhones 17 then perhaps? Nope, the Wi-Fi or cellular device you're using right now to read this article. Invention described down here...

In 4G and in 5G core networks we authenticate your User Equipment (UE) thoroughly before granting you a public IP address, because doing so makes the Mobile Network Operator (MNO) subject to regulatory liabilities : data retention, URL screening and lawful intercept, to name the most impacting. We grant you this UE IP address or prefix out of a pool (subnet) which each PDN Gateway (PGW) or User Plane Function (UPF) has announced to the datacenter switches and routers of the operator's network. The whole 4G/5G network is a giant IP-address-to-extended-radio-access-bearer mapping machine.

The more we want to deploy distributed UPFs, for example to reduce the latency for UE-to-UE communication or access to Edge Computing, the more we need to subnet the MNO's public IPv4 and IPv6 address space. The same occurs at a lower scale when we decompose a UPF into multiple Kubernetes Pods, for which we announce /24 and /56 micronets to the datacenter's leaf switches. When Kubernetes re-schedules the Pod in another (less loaded) worker node (host or VM), there you go, we need to re-announce that the SGi/N6 loopback address has moved via which those UE micronets can be reached.

As you move throughout the country, if we keep you anchored in the UPF of the city where you unlocked the phone in the morning, latency tends to increase, hence TCP or QUIC throughput is decreasing. 3GPP therefore specified (in 5G Stand-Alone) Session & Service Continuity modes allowing us to grant you a new IPv6 address at a new, nearby UPF. The cellular network became a bit more WiFi-like. Existing IPv6 flows are interrupted and the client must re-establish its TCP/TLS sessions from that new UE IPv6 address.

QUIC supports an advanced mechanism for connection migration to that new UE IPv6 address, including anti-spoofing, loss detection and congestion control. For a single app socket, QUIC can set up parallel streams over multiple IPv6 addresses, for example two 5G addresses and a Wi-Fi address. But a video stream is still delivered as chunks over the sequence-numbered streams, and just one stream per chunk, so when you lose contact to the old gNodeB, the video chunk must still be retransmitted via the new remaining UE IPv6 @. The impact is visible and latency remains high.

There's another nasty consequence of the regulatory burden and mandatory anchoring at cities having announced public IP subnets. International roaming and MVNOs remain home routed : all content is delivered via the gateways of the home operator or MVNO, not those of the visited MNO. 5G low latency communications (LLC) are incompatible with roaming and wholesale, the historic pillars of the success of cellular networks (!).

Therefore we at Nokia invented the 5G Ethernet PDU Type back in 2015, the EPS Ethernet PDN Type and the resulting R16 Ultra Reliable Low Latency Communications (URLLC). We no longer represent the UE only by an IP address, but also by a MAC address. LLC traffic can then be injected to a UE MAC@ in the visited PLMN or private 5G network, independently of home routing via the home IPv6 address. UEs can finally communicate to other UEs within the enterprise or visited country - even UEs with SIM cards from different operators. For that, the Edge Computing server, Mixed Reality box or router in front of it learns the UPF and Pod hosting each UE MAC@, for example by broadcasting ARP requests (IPv6 ND...) to all UPFs and Pods until the one serving the UE MAC@ responds. Any UE MAC@ could be behind any UPF and Pod. The SMF (and as of 3GPP R16 also the 4G/3G/2G SPGW-C) learns these UE MAC addresses by asking the UPF (or SPGW-U) to inspect the upstream frames popping out of the GTP-U tunnel. Next, the SMF/SPGW-C installs Packet Detection Rules (PDR) and Forwarding action Rules (FAR) connecting each MAC address to a GTP-U tunnel (hence radio access bearer).

In networks with many UEs and/or UPF Pods such ARPing is unsustainable : do not (mis)use your just-1-million-FIB-entry-routers (or worse, your datacenter switches) as mobility trackers. Do not flood your UPF Pods with ARP requests. And do not let your UPFs flood your routers with gratuitous ARP messages.

Another barrier is commercial : we are just launching the Ethernet PDU Type in our core networks this year; the supporting chipsets/modules/devices will have to follow and it could take a decade to reach critical mass.

Now when there's a techno barrier like this, at Nokia we solve it with what-ifs.

What if we could represent each UE with a virtual MAC address, carefully picked in a range for each UPF, UPF Pod, slice, subscriber group, tracking area etc? The SMF/SPGW-C could pick it and tell the UPF/SPGW-U, right? Well, that's possible when you set the 3GPP TS 29.244 PFCP (Packet Flow Control Prorocol)'s PDN Type to "Ethernet", even when the LTE/5G PDN/PDU Type indicates "IP" (v4/v6/v4v6). But you have a second hurdle : the uplink FAR cannot set the egress source MAC address representing the UE. Since with Ethernet PDU/PDN Sessions, the UE MAC addresses are already popping out of the ingress GTP-U tunnels (N3/S1-U).

I cracked that last nut by linking the UL FAR to the DL PDR using an intermediate, new, yet standard, N6-facing PFCP Endpoint ID. This allows the UL FAR to discover which UE MAC@ it needs to write in the upstream Ethernet frames. And of course no longer the burnt-in MAC address of the Intel or Mellanox NIC in the server hosting your UPF Pods, right... that was soo 2021.

I also filed the dependent claim to do this with Segment Routing labels rather than Ethernet MAC addresses.

Now what does this mean for #6G my friends? Well. Your serving 6GnodeB(s) can reach any other UE at neighbouring 6GnodeBs by knowing/learning the destination MAC@ range (or Segment Routing labels) of each neighbouring 6GnodeB. Just neighbouring 6gNodeBs or the whole network? What if we used L2 over stateless GRE over IPv6 to reach any? Oh wait, that would even work for 6GnodeBs of other operators or enterprises that are not even physically connected to you via fiber. Your could pass the frame or SR packet over the 6G air interface to another organization!!

In 6G there's thus no hierarchical mobility-tracking-home-routed core network as we know it in 5G/4G/3G/2G. Communication via internet peering points and rendez-vous points for internet apps becomes optional, auxilliary or deprecated. 6G is a flat, horizontal federation of peered #radiocore networks. A radiocore is a network mapping UE MAC addresses or SR segment IDs to PDCP sessions : there' s no more needless N3/S1-U GTP-U nor IPSec in the middle. Content is injected wirelessly. Two UEs in different radiocore nodes communicate via the most suitable path for each QoS : the shortest path for LLC but traffic-engineered, longer, segment-routed paths for less critical services. Wired inside the radiocore, wireless in between radiocores.

If home networks or MVNOs need to pay the visited MNO or ski resort per Megabyte, the visited radiocore needs to publish streaming user plane analytics on demand, providing confidence in the quality delivered to the roaming subscriber, I wrote about that last year.

Now there's nothing preventing us from representing 5G/4G/3G/2G UEs with MAC addresses in ranges that make sense, so why wait for 6G to flatten the network?

In my next post I'll keep you posted on the next steps which we have undertaken with our Major Customers.

#innovationpower