How to achieve end-to-end network slicing?

GPP defines network slicing as one of the main functions of 5G networks. Network slicing can be seen as a dynamically created logical end-to-end network. Before delving into the concept of network slicing, let’s briefly review the three major application scenarios of 5G.

5G use cases

3GPP, the main body for mobile specification development, is working to enable three fundamental use cases for 5G:

• eMBB (enhanced mobile broadband): refers to further improving user experience and other performance based on existing mobile broadband service scenarios, pursuing the ultimate communication experience between people.
• Ultra-Reliable Low-Latency Communications (URLLC): Communications with strict reliability and latency requirements for mission-critical communications, including autonomous vehicles, remote surgery, or the Tactile Internet.
• Massive Machine Type Communications (mMTC): The need to support communications among a very large number of devices in a limited area that can only send data intermittently, such as use cases related to the Internet of Things (IoT).

Figure 1 5G use cases

As shown in Figure 1, 5G networks must simultaneously support diverse and extreme requirements for latency, throughput, capacity, etc., and require a carefully designed architecture to provide the best balance between service capabilities and network investment. Operators should also commit to specific service level objectives (SLOs) to achieve their business goals or adhere to the agreed functionality for each use case. This is where network slicing comes in.

Defining network slices

3GPP defines network slicing as:

"A network slice is a logical network that provides specific network capabilities and network characteristics."

Ideally, network slicing allows the dynamic creation of logical networks within the same physical network to support different use cases and traffic loads. Network slicing is an end-to-end concept that extends from user equipment to the access network (AN), transport network (TN), and core network (CN).

End-to-end slicing provides the right isolation, resources, and optimized virtual network architecture to serve specific use cases, SLO requirements, or business solutions.

Network slices are orchestrated to form specific service logical networks running on the same physical network, which meet certain service attributes such as data speed, capacity, latency, reliability, availability, coverage and security. Network slices enable operators to establish different functions, deployments and architectures for each use case or service group, and can run multiple network instances in parallel.

Figure 2 End-to-end 5G slice scope

As shown in Figure 2, a typical 5G network can be summarized into the following parts:

• User Equipment (UE): An end-user terminal connected to a mobile network via the “air interface”.
• Access Network (AN): The set of elements that make up the Radio Access Network (RAN), including base stations, antennas, and spectrum resources.
• In the case of a disaggregated RAN, it also includes elements such as the Radio Unit (RU), Distributed Unit (DU), and Centralized Unit (CU). Figure 2 is a simplified diagram and does not show all possible RAN functional splits.
• Core Network (CN): A set of functions including signaling, authentication, user management, mobility, interfaces to external networks, and other control plane and management plane services. The 5G core may be distributed in different locations within the network.
• Transport Network (TN): Used to carry transport traffic between AN and CN. In the case of a decoupled RAN architecture, there will also be instances where the TN interconnects RAN components (such as RU, DU, and CU).

Network Slicing Use Case Examples

Example 1: Autonomous vehicle slicing requires an end-to-end network to provide capabilities such as data rate, reliability, latency, communication range, and speed for a specific slice instance serving a use case.

Example 2: IoT slices serving applications such as smart metering or wearable medical devices need to securely, efficiently, and cost-effectively support a large number of low-latency and high-density IoT devices.

Implementing network slicing

As mentioned above, 5G network slicing can be used to ensure end-to-end performance, as well as service and application requirements to meet customer expectations. To implement network slicing, each network segment (access network, transport network and core network) must be examined holistically. The life cycle of network slices needs to be orchestrated across the entire network.

Figure 3 illustrates the key elements involved in an end-to-end slicing implementation. This particular network provides network slicing services to three customers, tenants A, B, and C.

Figure 3 - End-to-end network slicing

Tenant A has three different slices, while tenants B and C have one slice each. Each slice is built as an end-to-end network slice, consisting of several sub-slices:

• A RAN (sub) slice
• Transport network (sub) slices that connect RAN slices to core network slices
• A core (sub) slice
• The second transmission network (sub) slice connected to the core network

Each slice above is lifecycle managed by a domain-specific orchestrator/controller, called the Network Slice Subnet Management Function (NSSMF) in 3GPP:

• RAN slices are managed by the RAN controller or RAN NSSMF
• Transport network slices are managed by the transport slice controller or transport NSSMF
• Core network slices are managed by the core controller or core NSSMF

The NSSMF has the specific domain knowledge required to implement sub-slices in this domain. The NSSMF is responsible for:

• Create slices
• Maintaining slices
• Terminate the slice when it is no longer needed

Implement a northbound interface that exposes an abstract view of the domain and allows NSMF to use slices (see below).

The highest level of this hierarchy contains an end-to-end network slice coordinator, which in 3GPP terms is the Network Slice Management Function (NSMF). The NSMF has the function of stitching sub-slices together to create end-to-end slices. The NSMF communicates with the NSSMF through their northbound interface to do this. In turn, it also exposes an abstract northbound interface to allow the use of its services to create end-to-end slices.

Each domain-specific subslice allocates or provides one or more of the following resource types, depending on its nature:

• Virtual and physical network functions
• spectrum
• bandwidth
• Transport layer connection model
• Enhanced services (such as network analysis and security services)
• Quality of Service (QoS) Profiles
• Application Features

For example, core network slices can allocate dedicated computing resources for signaling traffic. Transport network slices can use mechanisms to allocate network capacity to each slice.

Hard and soft slices

The level of sharing of network resources, “hard slicing and soft slicing”, depends on the service level objectives related to the network capacity.

The main difference between "hard slicing" and "soft slicing" is that hard slicing results in network resources being dedicated to one slice, while soft slicing allows for the use of shared resources.

Allocating dedicated, non-shared resources to each network slice instance guarantees the performance, availability, and reliability required by each application or customer. However, if these resources are not fully utilized, they cannot be used for other slices. Therefore, hard slicing may not be very cost-effective.

Soft slicing allows controlled over-subscription of transport resources, allowing network resources to be used more economically for high-capacity applications with looser constraints.

Transmission network slicing

The first half of this article introduces the definition of end-to-end slicing and how it is implemented. The second half will focus on transport network slicing and how it is implemented.

A transport network slice can be defined as a set of distinct connections between physical network functions (PNFs) and virtual network functions (VNFs). Such transport network slices have deterministic SLAs to achieve the end-to-end SLOs of a full end-to-end network slice. These SLOs include parameters such as QoS, availability, latency, and packet loss.

IP and optical transport networks have been using various network virtualization technologies to deliver virtual networks for many years. So what’s new when it comes to transport network slicing? The key requirements for transport networks are:

• New SLO types are associated with, for example, latency, which were not strictly required before. In addition, there is a need to ensure that SLOs are adhered to throughout the lifecycle of the service.
• Data plane technology can be extended to support fine-grained traffic engineering
• Use streaming telemetry to gain near real-time visibility into network status and performance
• Enhance network programmability using model-driven approaches, such as YANG models
• Centralized path computation is required, which in turn requires new network visibility mechanisms
• Closed feedback loop between network and control system
• An abstract API that allows the transport controller (NSSMF) to communicate with the peer-to-peer coordinator (NSMF)

To meet the SLA of end-to-end network slicing in the future, IP transport network slicing must meet several requirements:

Table 1 - IP Slicing Requirements

Table 2 provides a set of candidate technical solutions that can meet the requirements in Table 1.

Table 2 - IP Slicing Candidates

The functional set of Table 2 can be combined to build a transmission network, which forms a closed loop with the central controller, as shown in Figure 4.

Figure 4 - Closing the loop to achieve segmented routing transport network slicing

In this example, there is a closed feedback loop between the data plane network that implements the transport network slice and the controller that manages and/or orchestrates the network. In this network:

• Services need to be enabled between the base station (gNB/eNB) and the mobile gateway (MG).
• The service has certain SLOs, such as a maximum latency limit.
• There is one controller that can be used as both the SDN controller and the transport NSSMF.
• The network exposes its topology to the controller via BGP-LS.
• The network also transmits telemetry information to the controller so that the controller has an up-to-date understanding of the network. Telemetry information includes things like link utilization and latency.
• The controller uses the path computation engine to compute the paths between the edge routers where the gNB/eNB connects to the MG.
• The controller transmits the calculated path information to the edge router using PCE or BGP SR-policy; the edge router embeds this path information into the data packets associated with this service.
• Once a path is established, the controller continuously monitors the network to ensure that the SLO continues to be met. If the network condition of the path where the service is located deteriorates, the controller will reroute the service to another compliant path.
• The controller, in its role as transport NSSMF, also exposes APIs to NSMF so that NSMF can orchestrate end-to-end slices that include transport slice components.

Why use segment routing for slicing

Traffic engineering allows service providers to offer differentiated services and enhanced SLOs. However, when it comes to achieving more granular control over traffic routing, network operators are often stymied by scalability issues.

Current traffic engineering solutions in packet networks based on RSVP-TE only support coarse-level control. Attempts to apply RSVP-TE to design finer-grained service flows always fail due to scalability issues. Segment Routing is a new tunneling mode that can be used in conjunction with software-defined networking (SDN) applications to address the challenge of achieving good scalability and fine-grained control.

Unlike RSVP-TE and Label Distribution Protocol (LDP), segment routing does not require control plane signaling on a per-tunnel basis. It only requires the ingress edge router to maintain state for each service, removing the state management requirement from the intermediate and egress edge routers. This makes segment routing much more scalable than RSVP-TE while providing most of the same functionality.

While segment routing provides the ability to construct forwarding paths in the network, some abstract intelligence is required to instruct the ingress routers what paths to use in the network, and what services to use. This intelligence can be provided by an external traffic engineering controller that acts as a stateful active path computation element (PCE), providing end-to-end control of network resources based on real-time network status. This ensures that expensive wide area network (WAN) capacity is used efficiently and that the network can provide specific service requirements, such as disconnection, when needed, due to its network-wide visibility.

The use of a centralized controller also facilitates the use of SDN in WANs, providing a more flexible approach to networking by automatically creating and/or removing bandwidth available to specific services. This in turn allows the introduction of services such as bandwidth calendaring or bandwidth on demand.

Therefore, segment routing is an ideal technology for transport networks to realize the service capabilities required by network slicing.