The pros and cons of network slicing in 5G
Although network slicing is useful for certain applications, like IoT and mobile networks, it might not be necessary at this point, especially when considering its cost.
Evolution in the fifth-generation mobile network involves a lot of moving parts, but the critical piece that separates 5G from being just a different radio access network model is the concept of network slicing.
Network slicing applies virtualization principles to mobile networks, and many tout it as the launchpad for a new wave of software-defined networking (SDN) and network functions virtualization (NFV) deployment. But arguments about the necessity and value of network slicing exist on both sides.
Perhaps the best description of network slicing as a technology is horizontal virtualization. The basic strategy is to combine the following partitioned elements:
- the 5G New Radio (NR) network;
- SDN technology that segments the edge and core networks; and
- NFV architecture that virtualizes 5G control plane components.
This then creates multiple independent virtual networks -- or slices -- that allow mobile operators to separate users, devices and applications that require a different quality of service (QoS). Slices can also be used to give mobile virtual network operators their own virtual infrastructure, potentially improving MVNO-based services.
Horizontal slicing offers cross-application and cross-service segmentation of a network. Each horizontal slice has its own virtual resources highly independent of the resources used for other slices. This horizontal network slicing differs from current network segmentation -- or vertical slicing -- which doesn't partition resources as much as it allocates them based on the application or mission.
The business side of network slicing
The business-level challenge of network slicing derives from the difference between features that are useful in new services and features that are essential to justify the cost. Network slicing is surely useful for future 5G projects, like MVNO, IoT and other new mobile network applications, but it might not be necessary. The largest IoT application is a smart home or building, and we already have those today without any 5G features.
Similarly, we already have MVNOs, in addition to video delivery that requires good QoS. It's hard to argue that those things need network slicing or even 5G.
Without question, we need 5G for additional bandwidth per user and for more users per cell, but laying a 5G NR implementation on a legacy 4G Long Term Evolution infrastructure can meet those needs. This architecture -- called 5G non-standalone (NSA) -- is already defined and being deployed, but it doesn't include 5G Core features, like network slicing.
A network slicing implementation would presumably require a full implementation of 5G Core, which doesn't have final specifications yet. A complete 5G NR and Core implementation would require considerable investment and a technology refresh, which may be difficult to justify given the current limited business case for 5G.
Are new specifications necessary?
At the technical level, network slicing creates virtual networks, which are already widely used in cloud computing. In the 5G network slicing specification, slices are called network slice instances. These instances must contain all the logic and elements needed for the slice to operate independently.
The current 5G specifications provide possible implementation elements and use cases, but they don't yet specify exactly how network slicing would work. We know slices exist both in the radio network and in 5G Core, and the two may be connected or linked to create a slice that envelopes a complete 5G service experience.
But why do we need a new specification? Cloud computing uses virtual network services, and we have virtual networks even in the wide area with software-defined WAN. In addition, cloud computing has given us a wide variety of tools to deploy and manage distributed applications, and these tools can work for distributed service features, as well. Why not use cloud computing approaches for network slicing, as they are already proven in the real world?
One answer may be the difference between horizontal and vertical slicing. Cloud computing has almost universally adopted the resource pool model, which allocates shared resources to applications based on their quality of experience needs. This approach ensures optimized resource usage and easily controlled operations costs because you can manage resources as a whole.
But resource sharing can affect cross-application performance and create security problems if the resources aren't truly separate. Network slicing, in theory, isolates the slices fully for better security and QoS control, but it risks lower efficiency when using network and hosting resources, in addition to higher operations costs due to the need to manage the slices independently.
The differences in technical approaches to virtualization between network slicing and cloud computing make it difficult to build a common network and hosting facility that optimally supports both approaches. That means early applications would have to justify the cost of implementing network slicing -- and we've already noted that current video, IoT and mobile networking applications don't require network slicing.
Various studies -- mostly from vendors that would benefit from a full 5G implementation, including network slicing -- suggest new applications will demand 5G Core and network slicing within five to eight years, and they're likely correct. It's what happens in between that is worrying operators.
Timing for network slicing
In the end, the network slicing debate seems likely to come down to a matter of timing. Today's dominant 5G non-standard model doesn't include network slicing as a feature, so any application or service virtualization would have to be deployed based on cloud computing's vertical virtualization model or an NFV deployment.
The cloud model is currently far ahead of NFV, so virtualization with the 5G NSA deployments will likely progress with that model. This means network slicing would be retrofitted into 5G networks later, perhaps beyond 2025, when new applications actually demand the stringent separation of resources that network slicing provides.
Some of the transitional issues related to network slicing -- and even the business justifications for network slicing -- may fall away if the final specifications follow cloud principles closely. Many operators that have been pushing for a cloud-native NFV model hope this will be the case, but that outcome is doubtful because 5G Core cites NFV specifications. As a result, we can't truly balance the pros and cons of network slicing until we see more development in both the possible applications and the technical specifications for implementation.