Challenges in Creating a Robust and Scalable Network

One of the reasons why mobile network outages tend to occur on a large scale is the architecture of these mobile networks. The architecture of mobile networks are built around a single core network^※ that manages multiple base stations and devices. Nowadays, smartphones and mobile phones are widely used, and it is not unusual for a single person to own several devices. As a result, telecom operators are responsible for tens of millions of devices and if a major failure occurs in the core network, this can cause communication outages that affect the devices of millions or even tens of millions of users.

• ※Core Network
  A massive information system for a mobile network that accommodates over 100,000 base stations and over 10,000,000 devices. Core networks handle information on mobile lines and location information as the devices move, as well as authentication and billing data.

It is also a difficult problem to balance robustness and the demands for enabling new features expected in 6G. For example, in order to realize the technologies that are expected from future mobile networks, including autonomous driving, metaverse, and digital twin, it is necessary to enable (upgrade) certain functions such as “MEC (Multi-access Edge Computing)” and “Network Slicing”.

However, the structure of today’s mobile networks are becoming more complex, and enabling new functions requires incredibly complicated work. Such work could cause service outages, and this has become a challenge for when these new functions are enabled.

Comparison of Mobile Networks and EC Websites on the Internet

What kind of technology is required to manage the connections of many users while frequently enabling new functions in the core network without causing any outages? In this context, The Research Institute of Advanced Technology used the architecture of large-scale E-commerce (EC) websites on the Internet as a reference to research and develop a new mobile core network. On some of these websites, despite daily use by tens of millions of consumers, new functions are implemented at an extremely high frequency, and services are operated without the constant supervision of administrators. What are the differences between the system architectures of mobile networks and EC websites? Let’s start by explaining that.

Characteristics of Mobile Networks

In the current 5th-Generation Core Network (5GC) specifications developed by 3GPP (Third Generation Partnership Project)^※1, important signals, including those that control devices, are sent and received on stateful systems using a synchronized transmission method. A stateful system is a system in which the components hold a “state”, and the system behaves differently according to this state. In mobile networks, information on devices’ locations and the connection status correspond to the “states”. These states are synchronously maintained by a set of multiple functions that make up the core network. These functions are called Network Functions (NF).

• ※13GPP (Third Generation Partnership Project):An international standard developing body that creates protocols for mobile telecommunication systems.

One of the challenges in these stateful synchronized systems is their lack of scalability. For example, the Access and Mobility Management Function (AMF), which is responsible for the connection between the base station and the core network, is constantly connected to a communication protocol called Stream Control Transmission Protocol (SCTP)^※2 . If a failure occurs to the AMF for any reason, it will need to establish connections between the large number of base stations and the redundant AMF, and then rebuild the access information from the large number of devices from scratch. In addition, states have to be synchronously held between NFs that have been defined for each function. This entails the sending and receiving of a large volume of states between NFs. Due to this architecture, once a major failure has occurred, an extremely long amount of time is required to resolve it, and in some cases, it cannot be resolved regardless of the time spent on it.

• ※2SCTP: A computer networking communications protocol for the transport layer of the Internet protocol suite, used in IP networks.

Characteristics of EC websites

On the other hand, for many web applications represented by large-scale EC websites, the system is designed based on serverless processing that uses stateless and unsynchronized communications.

In a serverless system, detailed programs are allocated for processing each client request. Programs for processing each request are dynamically generated in response to the client requests, and when the processing has been completed, the program itself ends. This means that when there are no client requests, the computer resources are not used. These programs can perform stateless^※3 processing for each program by storing clients’ states on a database. In addition, each program is isolated to one request, so if a problem occurs during its execution, the failure is localized. Provided that the performance of the database is scalable, this type of system can flexibly respond to a large volume of requests by utilizing a cloud computing system. In fact, major EC websites handle events, such as unplanned large-scale sales, without problem.

• ※3Stateless: A method in which systems do not store information and data and outputs are determined based only on input contents.

The Research Institute of Advanced Technology’s Initiatives

Due to this, the Research Institute of Advanced Technology has been conducting R&D with the idea that if in a mobile network, a core network can be launched in a stateless mode for each request from devices, like on large-scale EC websites, we can achieve the creation of a robust and scalable core network.

Procedure-Based Core Network

The Research Institute of Advanced Technology has independently developed a ”procedure-based core network”, a core network design modeled after EC websites. While the 3GPP’s core network implements NFs for each function, the procedure-based core network divides the software into small parts for each procedure that corresponds to a request from a device. As a result, a software process implemented with core network functions is allocated to each request from the user’s device. Therefore, even if a particular process goes down for some reason, its impact can be localized to just the device that sent the request. Furthermore, by storing the states handled by core network functions into a database and realizing pseudo-statelessness, communications that were previously synchronous can now be done asynchronously. By providing these core network functions reactively, we can create a core network in which only the necessary amount of computing resources are consumed when they are needed.

Research Contents (1): Removing the endpoints of stateful protocols from core network functions

At 5G base stations, AMF and SCTP association^※4 have to be established and maintained.

• ※4SCTP association: A logical communication line provided to applications by SCTP.

However, in the concept of process-based core networks, maintaining SCTP association is not suitable. Therefore, the endpoints of SCTP association need to be removed from core network functions.

That being said, it is difficult to change the implementation of base stations that have already been widely deployed. In this research, the “N1N2 GW (gateway)”, which removes SCTP association endpoints, is installed between the base station and the core network instead of the AMF.

The N1N2 GW establishes SCTP association with the base station and has the function of relaying signals between the user device and the base station, as well as the core network. As a result, by installing the N1N2 GW, the endpoints of SCTP association were removed from core network functions.

＜SCTP endpoints in a standard 3GPP core network (between a base station and AMF＞

＜SCTP endpoints in a Per-UE core network (between a base station and N1N2 GW)＞

Research Contents (2): Removing management of UE’s state/context from core network functions

In order to make core network functions stateless, restoring the state and context of a targeted device from the received request message will be required.

In this research, a search was carried out within the database using a unique ID that can identify the device used by the user from the request message that was received, and the required information was collected and processed. After the processing was completed, the updated information was written back on the database, and appropriate processing according to the device’s state and context was achieved without storing the UE’s state and context in the core network functions.

Through the research that was described above in (1) and (2), it is possible to achieve a stateless core network. The creation of a core network that is robust and scalable can reduce the operating costs for operators, and contribute to the realization of a more stable mobile network.

Toward the Future

Current mobile networks are designed and operated to prevent the failure of its components. However, there is always a possibility that something can go wrong somewhere on the Internet. We aim to achieve carrier-grade quality and performance by conducting performance tuning^※5 in mobile systems based on the assumption that abnormal states may occur.

• ※5Performance tuning: Adjusting hardware, including CPU, memory, disks, and network, as well as resources such as OS resources, to ensure that they can be more effectively utilized, and to get the best out of the system’s performance.

The utilization of cloud computing is also being discussed for achieving a more robust and scalable mobile network. However, in order to fully leverage the benefits of cloud computing, it is important to redesign the core network to align with the unique characteristics of the cloud. Process-based core networks have a design concept similar to that of EC websites, which are already well-suited to the cloud, and are highly compatible with the cloud. We therefore intend to proactively pursue research on the implementation of these networks in the cloud.

At the Research Institute of Advanced Technology, we will work on the design and utilization of a new core network while continuing to conduct further research.