Cloudified reconstruction of telecommunications networks is already the consensus of global operators. Virtualization technology brings many advantages such as cost reduction and flexible scale-in/out. However, with the advent of 5G, new challenges and problems are constantly emerging.
5G has three major application scenarios: enhanced mobile broadband (eMBB), focusing on 4K/8K HD videos, VR/AR and other high-bandwidth services, requiring a transmission rate 10 times faster than 4G; high reliability and low latency communications (uRLLC), focusing on high-reliability and low-latency services such as self-driving car and telemedicine, requiring a delay as low as one millisecond; and massive machine type communications (mMTC), focusing on smart city, smart home and other massive connection services, requiring support for accessing one million devices per square kilometers, 10 times that of 4G. Therefore, 5G networks need to have higher performance and more powerful management capabilities, and also need to be flexible and intelligent to meet a wide variety of application scenarios. At present, the early NFV network infrastructure cloud solutions have bottlenecks in many aspects, and need to be further reformed in terms of deployment architecture, network performance, and operation and maintenance (O&M) convenience.
5G completely realizes control and user plane separation (CUPS), driving the network evolving to a distributed deployment architecture. On the one hand, flexible, high-performance edge nodes are built at the edge of the network to get close to end users: through the local offloading of high-bandwidth services such as 4K/8K and AR/VR, the occupation of the core network and backbone transmission network is reduced, the utilization of bandwidth resources is effectively increased. The high-speed processing capability is moved to the edge, effectively supporting services requiring ultra-low latency such as self-driving car and telemedicine. Edge nodes also need to be flexible and scalable to meet diverse 5G application scenarios. On the other hand, a network-wide intensive central node is constructed to provide a resource pool shared across regions, improve resource utilization, achieve efficient centralized management of massive nodes, and cope with rapid network development. The central node can also support a capability exposure platform to provide digital services and help operators achieve value innovation. Therefore, a distributed deployment architecture with features such as flexibility, high performance, and high efficiency will be the main development trend of telecom cloud network.
Early telecom network cloudification solutions turn traditional network infrastructure based on dedicated hardware into the unified resource pool based on common X86 servers, breaking down resource silos and achieving flexible resource scale-in/out. However, with the advent of 5G, facing the challenges from the performance requirement for ultra-low latency and exponential growth of service scale, common servers lose competitive edge in performance and cost. Hence, hardware acceleration technology is tightly concerned by the industry. Current mainstream hardware acceleration technologies involve offloading the data switching function of virtual switches on the telecom cloud platform to the FPGA SmartNIC to improve forwarding performance, and incorporating the graphics processing unit (GPU) into the unified cloud resource pool as a high-performance computing resource that can make full use of GPU’s excellent processing capabilities to improve computing capability of the telecom cloud platform and to provide better support for HD video and AR/VR services. ARM processor virtualization can also be used. The new-generation telecom network cloudification solution needs to converge these hardware acceleration technologies to comprehensively improve both forwarding and computing performance.
AI - Based Intelligent O&M
5G drives network functions (NF) moving to edge nodes close to end users. This trend leads to the dramatic growth of edge nodes by ten times even hundred times, and the O&M workload gets doubled accordingly. Besides, after the cloudified reconstruction of network structure, though layered decoupling significantly reduces hardware costs, more complicity is brought to O&M work. Therefore, telecom operators are looking for more efficient O&M approaches. To this end, deep applications of AI will become the key driver for automated and intelligent O&M. The AI technology is capable of analyzing multidimensional complicated problems across layers and domains, brings higher processing efficiency to multiple aspects of O&M which involve rapid root cause analysis (RCA), real-time dynamic resource adjustment, and capacity predication and analysis, and gradually enables fully automated O&M mode to release manpower and reduce Opex.
Container + Virtual Machine
5G NFs will be based on components and microservices. With less resource occupation and easy migration, container is considered as the resource carrier more fitting to 5G microservice architecture. However, the container technology has not yet matured in the telecom field, as it has weaknesses in orchestration capability and security. Moreover, most of current cloud projects use VM solutions which are constantly evolving in long-term practice and have many advantages. Thus, both the container and VM technologies are quite important to 5G, and the industry is also actively exploring how to choose and balance between them.
To address such challenges, ZTE has developed the 5G-ready 4MIX distributed cloud infrastructure solution (Fig. 1). It is based on the distributed architecture with “core cloud + edge cloud + access cloud”, and integrated with HCI, container, hardware acceleration, AI and other advanced technologies to build the 5G-ready cloud infrastructure featuring green, energy-saving, flexible adaption, performance acceleration, intelligence, and high efficiency.
The 4MIX distributed cloud infrastructure solution has the following features (Fig. 2):
- MIX deployment modes: The solution provides auxiliary deployment modes rapidly and accurately for different data centers, such as automated large-scale deployment for core cloud and green lightweight deployment for edge cloud.
- MIX resource pools: The solution combines two mainstream open source cloud platforms, OpenStack and Kubernetes, to build an integrated resource pool, to carry out unified management and orchestration of VM, bare metal, and container sources, and flexibly allocate resources according to upper layer applications.
- MIX hardware: The solution combines X86 servers and acceleration hardware such as FPGA properly and carries out unified management through the same cloud platform to significantly reduce hardware investment, guarantee high performance of the network, and finally offer the most cost-effective solution to users.
- MIX O&M mode: The solution builds an end-to-end closed-loop automatic O&M for the entire distributed cloud through remote control and AI technologies, bringing an efficient O&M mode with unmanned remote sites and centralized control at the center.
The 4MIX distributed cloud infrastructure solution creates the optimal configuration for excellent user experience and precise deployment to meet the requirements in different 5G scenarios, facilitating the construction of 5G-ready cloud infrastructure. The solution won the “Best New Cloud Infrastructure" award at the SDN NFV World Congress 2018. This award fully showcases ZTE's innovation capability and leading position in the SDN/NFV field.
4MIX distributed cloud solution, 5G-Ready cloud infrastructure, distributed deployment, container, virtual machine, hard acceleration