Where does the power of high-performance 5G core network come from?

The core network is the brain of the entire communication network and an indispensable and important component.

[[331159]]

The core network is mainly responsible for key functions such as network management and control, authentication and authorization. The strength of the core network directly affects the performance of the entire network.

In the 5G era, the entire mobile communication network architecture has undergone earth-shaking changes. In order to achieve ultra-high speed, ultra-low latency, and ultra-large connections, all aspects of the network have been transformed and innovated. So, in the face of stringent performance requirements, how does our 5G core network innovate itself and face challenges?

On June 19, at the Intel® Data Innovation Summit and New Product Launch Conference, Mr. Tu Jiashun, Chief Scientist of ZTE NFV/SDN, shared with us ZTE’s practical achievements in high-performance computing of 5G core networks.

Compared with 4G, the 5G core network is completely different in terms of service architecture and hardware platform.

By introducing NFV network element function virtualization technology, the traditional dedicated hardware platform is replaced by the x86 general hardware platform.

A large number of general-purpose servers form a hardware resource pool. On the resource pool, a large number of virtual machines and even containers are built through the virtualization software platform based on OpenStack. The network elements of our 5G core network are deployed on these virtual machines and containers in the form of software to realize the corresponding functions.

This kind of architecture is called microservice architecture, and this open platform is what we often call telecom cloud.

The virtualization and cloudification of the entire 5G core network make it highly flexible. Hardware resources are flexible, software deployment is flexible, and service migration and capacity expansion and contraction are also flexible.

In order to further reduce the complexity of operation and maintenance, the 5G core network also introduced an orchestrator.

The orchestrator is responsible for orchestrating resource scheduling. The orchestrator also integrates an artificial intelligence engine and even big data analysis to form some configuration libraries. The early experience and operation and maintenance requirements are transformed into a series of automated operation and maintenance policies, giving the entire system the ability to automatically deploy.

After adopting the microservice architecture, the efficiency of the 5G core network has been greatly improved, with deployment efficiency increased by 30% and operation and maintenance efficiency increased by 40%, which is a huge advantage compared to the 4G core network.

5G core network cannot be separated from efficient and powerful cloud infrastructure

In addition to the control function, one of the most important indicators of the 5G core network is forwarding efficiency.

The 5G core network needs to forward user data from the wireless access network to the Internet. Similarly, it also needs to forward data from the Internet to the access network.

The huge amount of traffic in the 5G network must all pass through the 5G core network, and the network element specifically responsible for forwarding is the UPF.

UPF is the key to 5G core network performance

Therefore, the performance of UPF directly determines the performance of the entire 5G core network and even the performance of the entire 5G network.

So, under the current architecture, how can the forwarding capability of UPF be improved?

According to Tu Jiashun, there are currently three main ways to improve the performance of 5G UPF network elements:

The first is CPU acceleration, which means that the UPF software is directly deployed on a more powerful CPU to improve performance.

The second is general network card acceleration. The network card is the direct entrance and exit of traffic and carries a large amount of data forwarding. Therefore, accelerating the network card can improve the performance of UPF.

The third is to use smart network cards. Smart network cards are special network cards that use some intelligent technologies to achieve direct data forwarding on the network card, thereby improving performance.

Here I would like to introduce Intel's SST-CP (Speed ​​Select - Core Power) technology.

SST-CP technology is a core frequency control technology supported by the second-generation Intel Xeon Scalable Processor N series. It can flexibly configure and adjust the priority of CPU cores and force some cores of a CPU to run at a higher frequency.

The UPF network element has high performance requirements, so the UPF is defined on a high-priority CPU core, thereby greatly improving the working performance of the UPF.

Why not just increase the frequency of all cores?

Of course, this is to save energy, and only network elements with performance requirements are allowed to work at high frequencies. 5G core network control network elements, such as AMF, do not require such high performance. If all are increased, the energy consumption of the entire CPU will inevitably increase, and the heat load of a single CPU will also exceed the standard.

SST-CP technology is a very practical technology that not only meets the performance requirements of UPF network elements, but also balances the overall power consumption of the 5G core network.

The second technology used to improve UPF performance is DDP (Dynamic Device Personalization).

DDP is a technology that makes network cards more "smart".

Traditionally, UPF, as a traffic channel, requires some CPU cores to distribute traffic and some CPU cores to forward traffic, with a ratio of about 1:2.

If the network card introduces DDP technology, it can download some User Profiles according to the needs, and then distribute the work according to the user's IP address as an index. In this way, the network card has the distribution capability, and one-third of the CPU cores used for distribution are released to do other work.

In other words, the introduction of DDP can theoretically improve system performance by one third.

The ZTE 5G core network test results shared by Tu Jiashun also fully verified the practical effectiveness of the two technologies.

System performance comparison before and after the introduction of SST-CP technology

As can be seen from the above figure, after the introduction of SST-CP technology, the overall performance has been improved by about 3%.

In fact, everyone knows that Intel CPU itself has a turbo frequency technology that can dynamically adjust the CPU's operating frequency, but this adjustment method is constantly changing, which will bring some instability and fluctuations to the overall function and performance. SST is a forced improvement, although the overall data does not look obvious, but the system stability and reliability are much better.

The effect of DDP is even more obvious, from 96G to 128G, it is almost 33%. This result is completely consistent with the previous theoretical analysis.

System performance comparison before and after the introduction of DDP technology

In general, the combination of SST and DDP technology can improve the overall performance of the system by 37%. This is a remarkable performance improvement for the entire 5G core network, and the power consumption and integration of the network will also be greatly improved.

DDP technology also brings a benefit, which is to improve latency.

As we all know, one of the major features of 5G is its ultra-low latency, which is mainly used to meet the needs of latency-sensitive scenarios such as Internet of Vehicles and industrial robots.

After the introduction of DDP technology, the network card directly distributes and forwards packets, reducing the upper-layer processing steps, which significantly improves latency. The test results show that the latency is reduced from 150us to 74us, almost halved.

Latency improvement effect brought by DDP

In the 5G era, performance is productivity, and high performance means high efficiency.

Network function virtualization (NFV) has laid a solid foundation for doubling the performance of the 5G core network. Hardware acceleration technology and hardware intelligence technology represented by SST and DDP have further explored the performance potential of the 5G core network.

With the development of the times, the performance of hardware will continue to improve, and the upgrade of architecture will be endless. It is these continuous innovations that have brought surging power to the progress of human communication!