As mobile internet and IoT develop rapidly, there are needs for diverse 5G services. 3GPP has defined three major 5G scenarios: enhanced mobile broadband (eMBB), ultra reliable low latency communications (uRLLC), and massive machine-type communications (mMTC). The eMBB scenario provides high-traffic mobile broadband services, such as high-speed download, HD video and VR/AR, with peak rates exceeding 10 Gbps and the bandwidth up to dozens of Gbps, which will put great pressure on wireless backhaul networks. Therefore, it is necessary to deploy services to the network edge as much as possible for local offload. The URLLC scenario provides ultra reliable and low latency communication services, such as automatic driving, industrial control and telemedicine, requiring an end-to-end high reliability of up to 99.999% and an end-to-end ultra-low latency of less than 1 ms to meet the higher requirements of the digital industry. In this case, services need to be moved to the network edge to reduce the latency caused by network transmission and multi-level service forwarding.
Traditional large-scale centralized telecom cloud cannot meet the requirements of 5G eMBB and URLLC scenarios. Multi-access edge computing (MEC) thus comes into being. As a key technology of 5G evolution, MEC provides cloud computing and IT service environments for edge applications at the network edge closer to customers. MEC features ultra-low latency, ultra-large bandwidth, and local real-time analysis and processing. On the one hand, MEC is deployed at the network edge, and edge services operate on terminals, resulting in faster feedback for addressing the latency issue. On the other hand, MEC allows contents and computing capability to be given at the edge and provides intelligent traffic scheduling, local offload and local content cache, so that some regional services can be terminated locally. This not only improves user experience but also reduces the occupation of backbone transmission and upper-layer core network resources. Therefore, MEC will be the best choice for 5G edge cloud deployment in the 5G era.
Location to Deploy MEC
MEC does not restrict its network deployment mode. It can be flexibly deployed in accordance with specific service scenarios and latency requirements. MEC resources can be deployed in access office (AO), central office (CO), and regional office (RO), as shown in Fig. 1.
By deploying a MEC platform at the edge, cloud computing is extended from the center to the edge, so that services can be rapidly processed and forwarded as close as possible to meet the need of diverse 5G application scenarios.
Way to Deploy MEC
Due to the limited environment (space, heat dissipation, and load bearing) of the edge site office and deployment costs, the hardware infrastructure of MEC usually uses universal servers based on x86 processors. These servers feature small size, low power consumption and high computing density, so they are suitable for deployment in edge offices that have lower environmental requirements.
MEC based on a specific cloud platform such as OpenStack provides a unified virtualized software environment and resource management for upper-layer MEC applications. The MEC applications are flexibly deployed as virtual machines on the edge cloud platform to meet the needs of different scenarios. For example, the applications strongly related to traffic forwarding are usually deployed on the edge DC, including vCPE, OLT-U, DU, BNG-U, UPF/GW-U and MEP, to meet the service requirements of high bandwidth and low latency.
ZTE MEC Deployment Solution
ZTE has proposed a complete solution for deploying 5G-oriented MEC to meet the needs of multi-service scenarios in the 5G era.
Deploying a Lightweight Edge Cloud Platform
As restricted by the environment of the edge site office, the deployment scale of MEC is usually small. The number of servers deployed on a single site is small, and the available hardware resources are limited. However, installing infrastructure platforms such as OpenStack and management modules may occupy a lot of resources, resulting in a waste of resources in the edge cloud. It is therefore necessary to downsize the MEC to achieve lightweight deployment, reduce the resources occupied by the platform and management part, and improve the resource utilization of the edge cloud. The downsizing measure involves:
—Tailoring OpenStack to retain only necessary components, remove redundant components, and reduce the occupation of computing and storage resources on the edge cloud.
—Adopting the dual-core (VM + container) solution for lightweight deployment of upper-layer applications. This facilitates fast deployment and upgrade.
—Combining thin OpenStack with the computing nodes on the edge cloud, so that physical resources are not exclusively occupied and the occupation of the control part can be reduced.
In the ZTE lightweight MEC deployment solution (Fig. 2), lightweight Openstack is deployed to reduce the number of components by 60%, necessary configuration resources by 75% and management resources by 10%. This not only greatly reduces resource requirements of MEC, but also significantly improves its management efficiency.
To improve the adaptability and resource density of MEC deployment, ZTE and Intel jointly launched a dedicated open edge platform (OEP600) that adopts a small all-in-one machine with the chassis depth of only 450 mm. Its CPU uses latest Intel's Xeon scalable processor to provide higher computing performance for edge computing. OEP600 supports wide temperature operation, strong heat dissipation, and easy maintenance, meets the requirements of multiple MEC deployment environments, and achieves the best match between performance and cost. At the MEC Technology and Industry Development Summit 2019, ZTE won the MEC Technology Innovation Award 2018–2019 for its edge computing hardware platform OEP600, fully demonstrating ZTE's innovation capability and leadership in the field of edge computing.
Deploying a Unified MEC Management Platform
Edge clouds are usually small in scale, large in number and dispersed in location, which brings great complexity to their planning, deployment, operation and maintenance (O&M). It is therefore necessary to deploy a unified MEC management platform on the upper-level convergence sites to manage lower-level edge sites in a unified manner (Fig. 2). Only computing nodes and storage nodes are deployed on each edge cloud to reduce resource usage of the management module.
The unified management of MEC is implemented in two aspects: resource management and O&M management.
—Unified resource management: The MEC management platform manages and allocates resource pools (computing, storage, and networks) on all edge nodes in a unified manner. It provides a unified interface to centrally monitor the topology, alarm, performance, capacity, and other information of physical resources on each edge node, and also provides fault location means such as log and alarm analysis for infrastructure administrators. NFVO, which is only deployed on the upper-layer convergence nodes, directly interconnects with the unified MEC management platform to avoid interconnection with the resource pools of all edge nodes and to uniformly orchestrate and deploy virtual machines (VMs) and containers deployed on all edge nodes.
—Unified O&M management: The MEC management platform provides unified O&M management for VIMs in each edge cloud, including site management, user/tenant management, feature configuration, image distribution, centralized backup, upgrade/patch management, inspection, and API distribution. It provides unified FCAPS management, unified alarm, configuration, and performance statistics. It also provides smart and simple automation tools for fast installation and upgrade, fast inspection, fast fault analysis and location, and log analysis to improve O&M efficiency.
Deploying 5G UPF to the Edge for Local Offload
To meet the requirements for big bandwidth and low latency in 5G application scenarios, MEC will be deeply integrated with 5G network architecture during deployment, and its service distribution, policy control, and QoS guarantee will be implemented through standard 5G network functions. Based on the C/U separation architecture of a 5G core network, user plane function (UPF) needs to be deployed at the network edge to reduce transmission latency and implement local offload of data traffic. The control plane's functional network elements such as SMF are deployed in the central DC for unified control of UPFs deployed in MEC as well as unified configuration and distribution of offload policies.
Deploying 5G UPF to the edge for local offload is shown in Fig. 3. The local MEC informs the policy control function (PCF) of the UPF offload rules. PCF configures the offload policy to the session management function (SMF) for centralized scheduling of all traffic. The local UPF can be selected through the local area data network (LADN), uplink classifier (UL-CL), and multi-homing for traffic offload. The local UPF can also send non-local traffic to the central UPF for processing. This prevents all traffic from bypassing the central network, reduces transmission pressure and network construction costs of the backbone network, and improves transport efficiency and user experience.
Providing Hardware Acceleration for MEC
MEC hardware generally uses X86 universal servers, but the X86 universal servers have low performance in processing specific service requirements, resulting in low performance-to-price ratio and failure to meet commercial deployment requirements in 5G scenarios. Different hardware and software acceleration solutions need to be used for different services:
—Computing intensive services such as 5G CU PDCP air interface encryption/decryption and MEC location algorithm that consume a lot of CPU resources need special hardware acceleration.
—Traffic forwarding services such as 5G UPF/GW-U, MEC local offload, CDN and BRAS-U that have high requirements for network forwarding capability need software and hardware acceleration for data forwarding.
—Video-related services such as AR/VR and video live broadcast need hardware acceleration for video rendering and transcoding.
—Training and reasoning operations involved in the AI field need the introduction of GPU for hardware acceleration.
ZTE hardware acceleration solution supports FPGA-based GTP acceleration, GPU-based video/audio acceleration, and QAT encryption/decryption acceleration. Especially for 5G UPF, hardware acceleration through a smart NIC is used to offload most traffic from CPU to the smart NIC for higher forwarding performance. At the MWC Mobile World Congress held in Shanghai in June 2019, ZTE successfully demonstrated its 5G UPF solution based on the edge hardware acceleration platform. Compared with the virtual UPF without acceleration, the solution reduces forwarding latency by 90%, increases throughput by 200%, and reduces power consumption by 55%, which better meets special requirements for forwarding capabilities of edge data centers in 5G URLLC and eMBB scenarios.
Opening 5G Network Service Capabilities
Service capability exposure is another feature of MEC applications. MEC deployed at the network edge can perceive and collect wireless network information in real time, and expose it to third-party applications, which can optimize service applications, enhance user experience, and achieve in-depth integration of the network and services.
The MEC capability exposure architecture is shown in Fig. 4. An ME platform (MEP) is introduced in the MEC architecture to expose capabilities of 5G network. Through the southbound interface, MEP obtains information about the lower-layer network such as UE real-time location, radio link quality and roaming status, packages the information into different service capabilities, such as LBS capability, radio network information service (RNIS) capability, QoS capability and bandwidth capability, and then exposes them to the upper-layer third-party applications via the unified northbound API interface. In this way, more value-added services are provided, and QoS can be improved. MEP can also feed back the perceived information about upper-layer application services, such as service duration, service period, and mobility mode, to the lower-layer network. By analyzing the information, the lower-layer network can further optimize its UE resource allocation (such as allocating appropriate bandwidth resources to VIP users) and session management.
Relying on the MEC platform, ZTE MEC capability exposure architecture opens network capabilities to third-party applications, provides differentiated user experience, taps network value, and increases value-added incomes for operators. The research and standardization of open interfaces also helps to accelerate the development and launch of innovative service applications and to build a good MEC industrial ecological chain.
ZTE 5G-oriented MEC solution integrates multiple software and hardware technologies such as platform design, virtualization, hardware acceleration and MEP with 5G network architecture to provide a lightweight MEC with unified management, high performance, flexible exposure. Applications, services and contents can be localized and deployed in a close and distributed manner, which to some extent meets the special service needs in 5G scenarios such as eMBB, uRLLC and mMTC, and enhances user experience. ZTE hopes to work with more industry partners to explore the cooperation model of edge cloud, build a 5G-oriented MEC ecosystem, completely promote the commercial use of MEC, and jointly promote the booming growth of 5G edge services.