5G network slicing is not a good reason to upgrade to SA

There have been several news items recently on the relatively slow progress towards standalone (SA) 5G core networks. Many operators appear in no hurry to upgrade their NSA 5G network, noting that there are some technical difficulties and that it is unclear whether SA will lead to any new revenue streams.

Those keen to see a transition – often manufacturers and similar suppliers – suggest that a key benefit is the ability to offer network slicing. I have written several times before on this, and Dean Bubley has done an excellent job in analysing the issues, but it felt worthwhile to summarise and re-examine some of the arguments.

Network slicing aims to deliver different quality of service (QoS) to different groups of subscribers by creating virtual networks with differing loading and possibly different features. So, for example, emergency service users might be on a network that had far fewer users/MHz of capacity meaning that congestion was less likely than on the public part of the network. Some protagonists have assumed, or at least implied, that a slice could offer perfect connectivity to the users paying for it.

There are many problems with this namely:

1.??? Coverage is the issue, not congestion.

2.??? Improving quality for some inevitably degrades it for others.

3.??? Better approaches are already available on 4G.

4.??? Multiple slices quickly become unmanageable.

The biggest issue is that for most users QoS is constrained much more by coverage than connectivity. Some time ago when I was at OpenSignal I ran some analysis that suggested that on average users had good connectivity (multi Mbits/s and less 100ms latency) around 90% of the time. Of the 10% they didn’t then 9/10th was due to coverage and 1/10th to congestion. So, simplistically, if a user was given a slice with a guarantee of no congestion then their QoS would rise from 90% to 91%. Now this is hugely user dependent. A user on a campus with good mobile coverage (eg a connected machine) may never experience coverage issues. Conversely, connectivity to trains may be worse that 90% because of the challenge of cuttings and tunnels. Remember, also, that most rural coverage is 4G and that it is unlikely 5G will reach 4G coverage levels for many years, if ever. So many users will not be able to gain access to a 5G SA slice for some of the time anyway. For most users a slice will not guarantee 100% connectivity and may hardly change their real-world experience.

The next issue is that improving the quality for some will inevitably degrade it for others. Delivering a high-quality slice, especially with ring-fenced radio resources, could mean that millions get a worse service. If just a small number of the “unsliced” churn elsewhere then the net result of slicing could be a large loss for an operator. If there are many slices then the overall loss of efficiency of use of the radio resource could also result in materially lower capacity across the network. Slicing would need to be charged at a high premium to compensate, but as we have seen above, for many slicing will hardly change real QoS.

Differential QoS is a problem that has already been solved. In the UK, the Emergency Services Network (ESN) running on EE’s 4G network, provides priority access for emergency use. Indeed, the GSM-R solution enabled priority and pre-emption, allowing higher tier users to grab resources from lower tier users when needed. Slicing may be a more elegant way of solving this problem and one that enables greater ring-fencing of data and more, but if there was demand for differential QoS it would likely have already emerged, as it has for the ESN.

Finally, slicing could quickly get out of hand. As others have pointed out, imagine that each car had separate slices, one for in-car entertainment, one for IoT connectivity and one for autonomous operation (not that autonomous cars need this…). That would be three slices. Now imagine each car manufacturer had a separate arrangement with its chosen operator. That could be 5-10 manufacturers/operator resulting in up to 30 slices. And that is just for vehicles. Across all applications there would be at least 10x more. Slices could rapidly become both unmanageable and unable to deliver better QoS.

Much as been made of private users such as oil refineries that might want a slice for their own operations. But more likely they will have their own private network and not a slice on a public network. There are some valid wide-area requirements for better QoS for emergency services, for some utilities such as power networks, and perhaps for railways (although they tend to self-provide). But these are already broadly catered for.

Of course, there will be some use cases for slicing. But it is hard to see how enough of a premium can be charged for them given the limited benefits they offer and the potential loss of revenues from unsliced users. Better to spend the money on improving network coverage and capacity which will deliver improved QoS to all.