Why is NFV spreading so rapidly under the 5G trend?

5G's high bandwidth, low latency, and large capacity access methods are destined to rely on a completely different professional equipment model than before, which also creates more directions and opportunities. NFV is one of them.

The biggest change of 5G is at the network level, which requires the underlying network architecture to form a network slice. Another major trend is edge computing. Both of these make NFV increasingly important, because traditional equipment can no longer adapt to the development of 5G networks. Therefore, in the 5G era, NFV is a "must have". As a participant and leader of NFV, China Mobile continues to make efforts in network applications and hardware acceleration to promote the rapid development of NFV.

01 In the 5G era, NFV is a "must have"

With the rise of the Internet/mobile Internet industry, major operators around the world are facing the impact of the wave of network transformation. NFV is becoming a must for operators' network transformation.

A NFV industry report released by Global Market Insights predicts that the NFV market will exceed $70 billion by 2024. Another study shows that the NFV market will grow to $19 billion between 2018 and 2022. Nearly 79% of telecom professionals see NFV as a key strategic priority in the next five years.

After the Ministry of Industry and Information Technology officially issued 5G licenses in June this year, the progress of related industries has been significantly accelerated. The arrival of 5G will undoubtedly accelerate the process of network virtualization, including the demand for NFV, which will become more urgent.

NFV, a tool hailed as a tool that can reduce capital expenditures while improving network performance and reliability, will usher in a new explosion in the 5G era.

Industry experts pointed out that "with the advent of 5G, network traffic is growing rapidly, and the network is in urgent need of transformation, because the flexibility of the existing network cannot meet the needs of the entire industry in the future. This industry demand is not only for telecom operators, but also for various enterprise networks, cloud, and OTT. The process of virtualization is bound to accelerate."

It is understood that the current NFV standards and open source codes are basically mature and available, ETSI's R2 version is basically commercially available, and the R3 version focuses on MANO management, SDN collaboration, cloud-native NFV, etc. and has been postponed to the end of the year; the ARM architecture is extending from the mobile end to the DC and network.

Although NFV has entered commercial use in some areas, the overall commercialization progress of NFV has not been as good as expected.

At the "2019 China SDN/NFV/AI Conference", Wei Leping, Executive Deputy Director of the Science and Technology Committee of the Ministry of Industry and Information Technology and Chairman of the SDN/NFV/AI Standard and Industry Promotion Committee, said bluntly, "At present, the industrial progress of NFV in my country is still relatively slow. The three major operators should be very determined and unhesitating to quickly use NFV, especially to use NFV by taking advantage of the opportunity of 5G. Because you will never mature if you don't use it. All technologies are gradually matured and implemented through applications."

In Wei Leping's view, three major reasons have limited the rapid development of NFV. "First, due to the conflict of interests and weak support from the industry chain, NFV has progressed more slowly than SDN. Second, layered decoupling and cross-vendor integration are still the goals, but considering the complexity and effectiveness, layered decoupling within the same vendor is still an early solution. In addition, NFV is mainly used in vEPC, vIMS and some vBRAS scenarios."

02 NFV hardware acceleration needs to focus on "three key points"

However, with the rapid development of 5G technology and the rise of edge businesses, more and more businesses have the need to be deployed at the edge. In various vertical fields such as smart manufacturing, smart cities, Internet of Vehicles, cloud gaming, AR/VR, etc., latency and bandwidth have become the core concerns of these edge businesses.

In order to meet the deployment requirements of diversified edge services in 5G uRLLC, eMBB and network slicing scenarios, and to achieve low-latency, high-reliability and high-traffic network functions under the limited resource environment of edge nodes, hardware acceleration technology came into being.

Wang Sheng from the Network and IT Technology Research Institute of China Mobile Research Institute believes that "the high reliability, low latency, and large traffic characteristics of 5G networks and the rise of edge computing services put forward higher requirements for future network computing and forwarding capabilities. After network function virtualization, network elements and VSWs are implemented in software, which consumes CPU to meet the computing forwarding costs and power consumption. Hardware acceleration solutions are needed to offload atomic functional units to hardware acceleration cards."

On the one hand, when power consumption/cost is constant, it is difficult to significantly increase the number of CPU cores, which limits performance improvement; on the other hand, from the perspective of product application, the CPU unit cost performance increases slowly, and the single-bit power consumption decreases slowly. The model of reducing the single-bit cost through CPU stacking is unsustainable. In addition, software acceleration in virtualization can only make up for the performance degradation caused by the I layer, and cannot accelerate business, and the acceleration effect is limited. Therefore, hardware acceleration is very necessary for operators.

In terms of hardware acceleration industry applications, Internet giants such as Microsoft, Amazon, Tencent, and Alibaba Cloud are already at the forefront.

Internationally, Microsoft Azure's hardware acceleration solution uses FPGA for network acceleration, covering bioinformatics, neural networks, search ranking and other areas, becoming an industry benchmark; Amazon AWS launched the FPGA acceleration service F1 at the end of 2016, as an enhanced EC2 instance provided to third parties for leasing.

Domestically, my country's Internet companies are not far behind. In 2017, Tencent Cloud launched China's first high-performance heterogeneous computing infrastructure - FPGA cloud service, which promoted FPGA services that only large companies could pay for long-term use to more companies through cloud services; Alibaba Cloud uses AMD and NIVIDIA GPUs to accelerate image processing, AI and other industry applications in terms of intelligent acceleration engines, and uses Intel and Xilinx FPGA chips to provide good high-speed and parallel performance support for the scientific computing industry.

It is understood that there are many choices of popular acceleration chips on the market. Acceleration chips embedded in network cards to form smart network cards are the mainstream form of acceleration cards. Among them, FPGA is relatively mature in the current industry and has high flexibility in on-site programming; NP and SoC have high cost performance, but the industry maturity needs to be improved; GPU's main advantage is complex image algorithm processing.

Because different acceleration hardware is used for different acceleration scenarios and the resources on the acceleration card are limited, there is currently no one-card-for-multiple-purposes product in the industry. Different acceleration hardware resource pools need to be built to accelerate different businesses.

Analyzing the current telecom network elements, the demand for network acceleration is more prominent, and the demand for computing acceleration and storage acceleration of new edge computing services such as AI is more prominent. Operators currently focus on three aspects: first, OVS acceleration, providing universal network forwarding acceleration capabilities for all network elements; second, network element specific function acceleration, such as GTP; third, GPU acceleration for AI and image processing in MEC.

At present, the telecom network element acceleration function is still in its early stages, focusing on the GTP function offloading of the forwarding plane network element, and the specific functions are not yet stable. Considering the industry situation and flexibility, it is recommended to adopt the FPGA solution, which can be flexibly modified and adjusted. After the acceleration function is stable, it is recommended to gradually solidify it into low-power, low-cost non-programmable chips such as NP and ASIC to improve the cost-effectiveness of hardware acceleration.

03 How does NFV spread like wildfire?

It is understood that China Mobile attaches great importance to NFV networks and is a major participant and leader in the NFV industry. As early as 2015, China Mobile established the OPEN NFV laboratory and introduced 15 cooperative manufacturers to conduct three-layer decoupling tests and build a TIC platform.

In July 2015, China Mobile officially released its next-generation innovative network, NovoNet2020, to the industry, hoping to integrate various latest technologies to build a new generation of flexible network to meet the digital needs of the majority of users and to build the basic network for the development of the Internet.

The future network transformation represented by NFV technology is a complex system engineering that requires a new research and exploration model. China Mobile adopts a top-level design approach, combining key technology research with experiments and pilots to iteratively promote the transformation and development of future networks.

In terms of top-level design, in addition to the target network architecture, vIMS, cellular Internet of Things, fixed-line broadband access, CDN and other services are used as entry points for the introduction of NFV, and gradually expanded to all business areas; in terms of pilot projects and experiments, the maturity of NFV technology is verified by second-layer decoupling, and the operation and maintenance management experience after the introduction of NFV is explored, and gradually deepened into the experimental verification of third-layer decoupling; in terms of key technology research, phased progress has been made in NFV hardware selection, virtual layer technology requirements, MANO architecture, network management and operation and maintenance, etc.

In addition, China Mobile has further clarified the NFV hardware acceleration solution, and by working with the industry to promote the maturity and implementation of NFV hardware acceleration, it will accelerate the commercialization of 5G URLLC and eMBB.

Currently, NFV acceleration management and orchestration of virtual machines requires the collaboration of MANO, OpenStack, and accelerators to jointly open up the management and orchestration process.

When it comes to accelerating general APIs and decoupling methods, experts recommend that in terms of OVS hardware acceleration unloading, it is recommended to adopt a decoupling solution between OVS and hypervisor, with the acceleration network card manufacturer providing the OVS function and the virtual layer manufacturer integrating the acceleration network card OVS. In terms of network element hardware acceleration, there are two decoupling solutions for network element service acceleration: white cards and functional cards. Functional cards are used for solutions that can be standardized by operators, and white cards are used for solutions with large differences in vendor products. In the future, operators can consider solutions to customize acceleration network cards, gradually realize the convergence of acceleration network cards, and reduce the complexity of network element development and acceleration network card standardization.

At the Mobile World Congress (MWC) Shanghai, China Mobile Research Institute demonstrated the world's first 5G-oriented edge open hardware acceleration platform to the industry and released the "China Mobile Forwarding Plane Network Element Hardware Acceleration White Paper". It is understood that China Mobile Research Institute, together with ZTE, Lenovo, and Keysight Technologies, completed the first demonstration of this solution. In this solution, ZTE's virtualized UPF can call Lenovo's acceleration card, ensuring the versatility of the acceleration hardware, breaking through the current performance and latency bottlenecks of virtualized forwarding, and achieving ultra-high throughput and ultra-low latency of the virtualized forwarding plane.

04 Conclusion

The arrival of 5G will undoubtedly accelerate the process of network virtualization, including the more urgent demand for NFV. In the opinion of experts, the current application of NFV still has a lot of room for improvement. In addition to promoting the implementation of NFV in network applications, China Mobile advocates that the industry jointly conduct in-depth research and practice on hardware acceleration technology, especially the solution, architecture, and evolution path of hardware acceleration technology for forwarding plane network elements, to jointly promote the maturity of hardware acceleration technology and better support the business development and network transformation of 5G edge scenarios.

[This article is an original article by 51CTO columnist "Mobile Labs". Please contact the original author for reprinting.]

Click here to read more articles by this author