A major challenge in future mobile networks is how to fulfil the requirements set for 5G networks in terms of latency and throughput. 3GPP has defined a new architecture based on virtualization and Software Defined Networks (SDN) supporting network slices that can fulfil those requirements. In this paper, we present the first realization of the new 5G user plane function (UPF) component that supports SDN and provides optimized transport for reducing latency as required in 5G networks. The proposed UPF is the cornerstone for using different data transport strategies adapted to the needs of different verticals, such as Ultra Reliable Low Latency Communications (URLLC) services that require a separate network slice providingan optimized transport for URLLC applications. The paper also discusses how to best migrate from legacy 4G user plane to 5G UPF, so as to ensure a reasonable transition to 5G. In doing so, the objective is not to meet short-term needs but to fulfill the future latency and throughput requirements of emerging applications, and also to provide first performance results of UPF in a realistictestbed environment.
5G has set ambitious requirements in terms of latency and bandwidth on mobile networks. In the coming years, traffic demand and the variety in services supported by mobile networks are expected to increase, not only with the usage of high resolutionvideo (4K, 8K video streaming) but also due to the higher number of connected IoT devices. The current mobile networks rely on IP/GTP network shaping. However, thecurrent protocol stackbased on IP/GTP is not effective for avoiding network congestion or at least isolating the congestion to ensure reliability. In particular, the traffic prioritization at the IP layer is not sufficient to guarantee the level of reliability required by industrial or mission critical applications. Thus, the IP/GTP traffic shaping is inadequate and might not work properly when the network is congested.
Applications with low latency requirements require guaranteed service provisioning regardless of the network congestion and even under unexpected traffic load. Thus, a straightforward solution would be to reserve a physical link connection (OSI Layer 1), or to reserve links at Layer 2. This means that logical links at L1/L2 are reserved forthetraffic flows that requirelow latency. However, this approach does not lead to an efficient utilization of available network resources. The L1/L2 link reservation will result in having idle resourceswhen the traffic sent is below the allocated capacity. Asolution in which L2 links are statically allocated to the services that require lowlatency (without having any constant demand) leads to resource under-utilization. Therefore, a dynamic allocation of L2 links is a desirable functionality for a mobile network. Software Defined Networking (SDN) provides the capability to dynamically change traffic flow descriptions, via various approaches, such as modifying the number of L2 links based on the number of traffic flows that require low latency. This approach allows a traffic flow to continue to receive the best possibleservice at the GTP/IP layer, but through SDN dynamic allocation of L1/L2 links will guarantee the reliability and low latency for certain flows.
In this paper, we present an SDN-based approach to the user plane traffic flows between the Radio Access Network (RAN) and the mobilecore. The L2 links are realized either using physical or virtual ports in a physical or virtual switch that implementsa programmable 5G User Plane Function (UPF). We have deployed L2 links as virtual connections using Ethernet VLANs. The user plane traffic flows coming from the base stations over GTP/IP are terminated in the UPF which then proxies the traffic to the next switch using dedicated VLANs that resemble the reserved L2 links required by low latency applications. The UPF can then apply different actions or priority to the traffic flows on those VLANs based on the latency requirements. SDN concepts are used to set traffic flow rules into the UPF switch, to support dynamic traffic differentiation attheuser plane.
The result of the proposed approach is that network slices based on dedicated L1/L2 links can be implemented usingtheUPF that replaces GTP/IP with VLAN tunnels. SDN is usedto enforce fine-grained traffic management in the VLANs compared to L3/L4 traffic throttlingat IP/GTP layer. This solution creates the basis for 4G/5G network slicing.
In this work, we have deployed the first release of 3GPP-defined UPF in two different testbed environments in order to collect measurements from different infrastructures. The first measurements are obtained from the testbed in 5G Innovation Centre (5GIC)  in University of Surrey, and are conducted as part of EU SoftFIRE project . The second set of measurements are obtained from the testbed in AALTO University  as part of the TAKE5 project .
The results from both deployments show the benefit of deploying UPF as a stand-alone component performing local breakout for user data planerequired for Mobile Edge Computing (MEC) in mobile networks. Performance results show that UPF allows effective implementation of network slicing which is highly desirable to support URLLC communications.The rest of the paper is structured as follows. Section II describes the 4G and 5G architectures and presents the issues with current deployments. Then, Section III introduces the proposed solution to overcome the limitations encountered with the 4G architecture when using virtualization techniques. Section IV presents the testbeds that are used to collect the measurements from deploying the proposed solution in a realinfrastructure. Finally, conclusions are presented in Section V.
-  5G Innovation Centre, University of Surrey, https://www.surrey.ac.uk/5gic
-  EU SoftFIRE project, https://www.softfire.eu
-  Depart ment of Communications and Networking, Aalto University, Finland, http://comnet.aalto.fi/en/
-  Take5 Project, http://5gtnf.fi/projects/take-5/ . www.take-5g.org
-  Open Networking Foundation (ONF), https://www.opennetworking.org/
-  ETSI NFV Management and Orchest ration (MANO)
-  “Network Functions Virtualisation — Introductory White Paper”, ETSI, 22 October 2012, retrieved 20 June 2013.
-  OpenBaton, https://openbaton.github.io/
-  Intel 40Gbs (Ethernet Converged Network Adapter XL710 – QDA2)