We are pleased to share with you all an interesting article contributed by Thierry Van de Velde who is technology specialist in Mobile Internet networks, architecture and solutions.
Thierry Van de Velde
Consulting Technology Specialist at Nokia, IP & Optical Networking
Following last week’s blog entry expressing my critical opinion on Huawei’s Application-Dependent Networking  I was invited to provide my view on the core network architecture for 5G.
First of all a Next-Generation Mobile Core (NGMC) should not be designed to support only the 5G Radio Access Network (5G RAN). Korea Telecom’s 5G strategy  is for example clearly indicating that KT’s Trusted Wi-Fi Access Points (AP) should be interconnected to the same convergent core network. Not only to bring reliable deep indoor coverage and extra capacity for data and voice services, but also to ensure a robust and consistent end user experience while the UE is accessing Wi-Fi and 4G/5G concurrently.
Similarly for the Cellular Internet of Things (3GPP R13 CIoT) and the non-cellular Internet of Things (LoRaWAN, Wi-Fi, Bluetooth, ...) a common core network is desirable.
In today’s 3GPP Release 12 architecture  any Trusted non-3GPP access is handled by a TWAP (Trusted Wireless Authentication Proxy) and TWAG (Trusted Wireless Access Gateway) interconnecting the TWAN (Trusted Wireless Access Network) to the EPC (Evolved Packet Core) – in practice to the PGW (PDN Gateway) via the S2a reference point (GTPv2). Some feedback we got for that 3GPP R12 eSAMOG architecture is that today’s PGW are not dimensioned to accommodate Wi-Fi traffic (user plane and highly chatty control plane) and that charging or policy control could also be performed by the TWAG itself rather than by the PGW. Which is why we built Diameter Gx and Gy into Alcatel-Lucent’s 7750 SR WLAN GW (TWAG/TWAP) this year.
In the NGMC an AGC (Access Gateway Controller) would be the convergent control plane node (replacing the TWAP) and distributed Access Gateways (AGW) would be deployed at the Edge of the network (replacing the TWAG). Together they would handle 5G, augmented 4G+ and non-3GPP access as shown on the following diagram :
The main objective of decentralizing the AGW to the edge is to achieve ultra-short round-trip delays, if possible 1ms, for new use cases such as rapid vehicle-to-vehicle communication, virtual/augmented realty, piloting drones or ultra-responsive access to decentralized content (virtualized Content Delivery Networks).
The AGC and AGW fundamentally differ from the MME and SGW in that they would :
The 5G RAN and augmented R15 E-UTRAN would connect to the AGC/AGW via the F1 reference point. F1-C would be based on EAP/NAS/F1AP/SCTP/IP. And F1-U on L2/GRE/IP, where each UE is represented by a temporary MAC address derived from its GUTI (Globally Unique Temporary Identity). That temporary cellular MAC address (cMAC@) would completely replace today’s GTP-U TEIDs (Tunnel End-Point Identifiers on the S1-U reference point).
The AGC/AGW could partition a 5G RAN or any non-3GPP access network’s control plane and user plane resources among multiple EPCs, TDFs or PDNs, hence address the NGMN Alliance’s requirement for core network slicing . Each EPC could be dedicated to a particular service (APN), PLMN ID (MOCN) or device category (DECOR). 3GPP would define new policies which the PCRF would communicate to the AGC via Fx. The difficult tasks of resource partitioning and network slicing would thus not be thrown over the fence to the NFV Orchestrator or VNF Manager .
As in 3GPP R13 the TDF would analyze traffic at L3, L4 or L7 and then steer upstream and downstream traffic through Service Function Chains (SFC). Each SFC consists of L2 and L3 non-virtualized or Virtualized Service Functions (VSF), interconnected through SDN technology: the SDN controller sends OpenFlow commands creating flow table entries in the underlying Virtualized Routing & Switching (VRS) layer. A single UE could access different SFC based on the L3/L4 services and L7 applications it’s accessing.
Also here 3GPP could enrich the existing Sd reference point (on which today only per-UE dialogues are established) with policies protecting each SFC or VSF from congestion.
With this 3GPP R15 NGMC architecture and its proposed interworking with the R8-R14 EPC, 5G users could continue to be anchored in the EPC (MME, SGW, PGW), and/or in the TDF, or only in the AGW. In the latter case (via FGi) the UE would benefit from seamless handover between 5G, 4G+ (R15 LTE) and Trusted non-3GPP access, but not to unmodified (pre-R15) LTE, nor 3G, nor 2G, nor the Circuit-Switched domain. If a 5G UE needs CSFB or SRVCC it would continue to be anchored in the EPC. The S1’ reference point permits doing so : the AGC/AGW would appear as a 4G Small Cells Gateway to the MME/SGW.
In fact the AGC/AGW could be used to “stabilize” an EPC or to expand its capacity and lifetime. 70% of today’s MME and SGW control plane traffic is due to the continuous toggling between ECM-Active and ECM-Idle mode – deleting and restoring S1-U bearers every couple of seconds due to inactivity. If the AGC/AGW would maintain the S1’-U bearers while Radio Access Bearers (RAB) are deleted in the R15 E-UTRAN, existing unmodified EPC could accommodate a much larger number of UE. MME Paging would no longer be required as long as the UE is in 5G or 4G+ coverage.
Since 5G will appear in pockets of coverage the architecture should be designed to achieve a smooth transition and seamless handovers to existing unmodified 4G, 3G and perhaps even 2G. When LTE had been introduced in Release 8 (2008) the 3GPP had catered for smooth interworking with unmodified 2G/3G SGSN by adding an Annex D to its 23.401 specification (GTPv1-C interworking at the PGW). Back in Release 8 it was the GGSN which was suppressed from the EPC; this time in the Release 15 NGMC it would be the TWAG/TWAP which would give way to a more thoroughly specified AGC/AGW.
Since Alcatel-Lucent is in the process of being acquired by Nokia this article expresses my personal view and findings, based on my 18-year experience in packet-switched mobile networks.
 On Resource Partitioning, Network Slicing and Service Chaining, http://www.netmanias.com/en/?m=view&id=blog&no=8366
So the idea of ALU is moving all the user plane to the edge to shorten the latency supporting various applications by introducing AGW at the edge and AGW at data center to handle the control plane to MME. So what is the cost of doing this?
There are 2 schools of thought at present, the first is to deploy a "mobile computing edge" : one or two x86 servers at each aggregation or edge router location. The market price for such server is around 5K USD, and it can host multiple virtual machines such as vRAN (PDCP-LL or PDCP-LL+RLC or PDCP-LL+RLC+MAC) and vAGW; Docker containers for vCDN, VR applications; ... Other operators are interested to integrate the AGW in an aggregation or edge router with internal computing blade such as the new 7750 SR-e (https://www.alcatel-lucent.com/products/7750-SR-e) with Integrated Services Appliance (ISA). The ISA cannot host virtual machines, its price is about 50% higher than an x86 server, but throughput would also be higher and more predictable (25 Gbps of encrypted traffic).
Thank you Thierry for sharing your valuable view. It's obvious that the increase in UE/traffic will lead to an increase in network and core elements. As a matter of fact, NFV is one way to lower the cost, at the cost of specialized HW integrated/optimized functionalities. Hard call.
Thank you thierry. i think increasing Nodes in core may make complexity in both PNF and VNF area. also for handover it will make some problems. i think it's better to keep 5G contextes on EPC and let non-3gPP to use new nodes integrated with epc.