Network Slicing: A Booster for 5G

Preface

I have recently become interested in 5G network slicing. After reading some literature, I found a very good paper, "Reshaping the mobile core network via function decomposition and network slicing for the 5G Era", which provides a very comprehensive and detailed introduction to 5G slicing. I decided to extract some of the parts that benefited me, and then combine my own understanding to give a brief introduction and summary of 5G slicing. This can be regarded as a small progress for me, and I hope it can be of some help to beginners.

[[185646]]

5G, which will be commercialized in 2020, has been widely mentioned. When mentioning 5G, we have to mention network slicing. As the most discussed technology in 5G, network slicing is of great significance to 5G. This article briefly introduces slicing technology from the following aspects.

1. Why should we evolve to the 5G era?

Only by understanding the limitations of the traditional core network can we deeply understand why we need to evolve to the 5G era and further understand the necessity of network slicing technology for 5G.

First, with the rapid increase in the number and types of user terminals, the massive growth in traffic, and the increasing diversification of user needs, the current core network EPC (Evolved Packet Core) traditional centralized network architecture is gradually becoming unable to handle the increasingly diverse service requirements.

Second, EPC is a "one size fits all" architecture, which is "innately deficient". For example, in EPC, the main function of the Mobility Management Entity (MME) is to manage the mobility of terminals, but not all user devices are mobile. For example, the communication between machine-to-machine (M2M) sensors does not need to provide mobility support because the geographical locations of these devices are almost unchanged. The architecture of the traditional core network will make many of the original designs completely useless when facing specific user groups.

Third, many network elements on traditional core networks run on dedicated hardware devices and are heavily coupled with software elements, which is not conducive to network programmability.

Fourth, since the functional division of each component in the current mobile core network is not clear, many user packets will be processed repeatedly in the process from eNodeB to SGW and then to PGW, and the packet processing process is not simple.

***, due to its centralized architecture and high requirements for software and hardware, its deployment period is long and the cost is high.

2. Why should network slicing technology be applied in 5G core network architecture?

In the 5G era, the objects of mobile network services are no longer just mobile phones, but various types of devices, such as mobile phones, tablets, fixed sensors, vehicles, etc. The application scenarios are also diverse, such as mobile broadband, large-scale Internet, mission-critical Internet, etc. The requirements that need to be met are also diverse, such as mobility, security, latency, reliability, etc. This provides a place for network slicing. Through network slicing technology, multiple logical networks are divided on an independent physical network, thereby avoiding the construction of a dedicated physical network for each service, which is very cost-effective! Looking at the network slices divided according to the different needs of different groups described in Figure 1, it is easy to understand the value of network slicing! From the perspective of slicing, the traditional EPC can be regarded as a large slice that serves all supported mobile devices, but as mentioned earlier, using a unified network architecture to meet all service requests at the same time is not efficient and easy to implement. Therefore, the future network must transition from "one size fits all" to "one size per service" through network slicing technology.

3. Definition of Network Slicing

Network slicing can be understood as a set of logical network functions that support the communication service requirements of a specific usage scenario or business model. It is the realization of services based on physical infrastructure. These logical network functions can be regarded as a series of sub-functions (Network sub-Function) decomposed from the network function (Network Function) under EPC. It can be seen that network slicing is an end-to-end solution, which can be applied not only to the core network, but also to the radio access network RAN.

Network slicing considers issues from the perspective of the service layer and the infrastructure layer. The service layer describes the system architecture from a logical level, which consists of network functions and the connections between functions. These network functions are usually defined in the form of software packages, which provide templates that define deployment and operation requirements (connections, interfaces, KPI requirements, etc.). The infrastructure layer describes the network elements and resources required to maintain the operation of a network slice from a physical level, including computing resources (such as IT servers in data centers) and network resources (such as aggregation switches, edge routers, cables, etc.).

After dividing the infrastructure layer and the service layer, we need to consider the mapping problem between the two layers. This is a typical virtual network embedding problem, which mainly includes the following two steps: 1. Mapping from virtual functions to physical functions, including the selection of network forwarding elements and computing resources, such as the selection of device types and device geographical locations. The amount of resources required is determined by the needs of the service layer. 2. Mapping from virtual links to physical links. How much physical link bandwidth is allocated also depends on the needs of the service layer.

The relationship between any two slices A and B can be one of the following:

• Different service layers: For example, slice A provides services for M2M type devices, and slice B provides services for manually operated devices.
• Same service layer, but with slightly modified network functionality: For example, slice A supports highly mobile users, and slice B provides the same service but without mobility support
• Same service layer, but different physical applications: For example, slice A and slice B both support the same service, but A provides ultra-high reliability and B provides standard reliability, so the deployment requirements of the two slices are different.
• Same service layer, same physical application: for example, slices provided by different operators.

4. How to decompose network functions (Network Function → Network sub-Function)

The definition of network slicing is introduced above. It can be known that the definition of network slicing is inseparable from the division of core network functions. Functional decomposition refers to the decomposition of tightly coupled network functions on some network entities (such as MME, SGW, PGW, etc.) so that they can run on different network entities.

Currently, network entities in EPC can be divided into two categories: 1. Entities such as MME that are used for control tasks and therefore only handle control plane (C-plane) traffic; 2. Entities such as SGW and PGW that also need to handle user plane (U-plane) traffic. Currently, the decomposition of network functions is still in the research stage, and one of the divisions is provided here, as shown in Figures 2 and 3.

In this decomposition, EPC functional decomposition is defined into two categories: horizontal decomposition and vertical decomposition. Horizontal decomposition means decoupling control plane and user plane functions, and vertical decomposition means identifying individual functions contained in one network entity. After decomposition, combined with network function virtualization (NFV) technology, these decomposed sub-functions can be selected and applied by dedicated user groups.

5. Impact of application scenarios on system design

Although the use of network slicing can greatly improve network performance and service quality, the number of network slices is an important issue. We cannot set different slices for each service. Too many slices will make maintenance and management difficult, and too few slices may result in a simple use case requiring two or more slices to meet its requirements. The use scenario and performance requirements have a great impact on the design of slices. 3GPP and NGMN have also been working hard in this regard and have made some progress. Table 1 shows the impact of some application scenarios divided by NGMN on slice design. The table summarizes most application scenarios and classifies them into eight series. The degree of requirements for different services in each application scenario is expressed by a score from 0 to 3, with 3 indicating the most stringent requirements and 0 indicating the most stringent requirements. It can also be seen from the table that an application scenario may use more than one network slice.

6. What challenges does network slicing face?

Network slicing structure: Although eight application scenario series are well defined in Table 1, there are still many scenarios that have not been classified. Therefore, how to determine the granularity of slicing division in terms of performance evaluation criteria remains a problem that needs to be solved.

1. Network slice selection: A user may use one or more slices. How to choose the appropriate slice is also a basic problem.
2. Network slice conversion: In roaming scenarios, if the local network slice cannot support user access to the network, it will cause network interruption for the user. A possible solution is to convert the user to the default slice. However, how to maintain the connectivity of the IP session during the slice conversion process and whether the task of detecting the conversion timing should be left to the user terminal or the network are issues that need to be resolved.
3. User status maintenance: User status information may be transmitted in multiple slices, and how to manage user status is also a key issue.
4. Determination of new functions: In order to support some new services such as driverless driving, the current EPC functions may not be sufficient, so new functions and the message formats and processing procedures involved need to be defined.

About the author: Li Zishu, first year graduate student in CCN and NDN at Future Network Theory and Application Laboratory, Beijing University of Posts and Telecommunications