5G – Part 2 – Some Thoughts On Network Slicing

One of the new features of 5G networks is ‘Network Slicing’. So far I couldn’t really get my head around the concept, everything I heard was just too abstract to make practical sense. Finally, however, things are getting more concrete and 3GPP TS 38.300 has an interesting section of how network slicing will work on the radio network side of 5G. Let’s have a closer look.

Network Slicing in 4G LTE Networks

Let get on this from a practical point of view and by looking at 4G fist rather than 5G: The 4G LTE radio network was designed to provide wireless broadband Internet connectivity. The air interface was optimized for this purpose and does a great job at this. As a consequence, however, it is not very suitable for very low data rate communication and especially not for devices that should consume as little power as possible. This resulted in the specification of the Narrow-band Internet of Things (NB-IoT) air interface. NB-IoT can be deployed inside (inband) of an LTE carrier, in its guard band or stand-alone. Let’s think about the inband option for a moment: In effect, it means that some resources in the frequency and time domain are taken away from the LTE resource grid and used for NB-IoT. The LTE scheduler never assigns those resources to LTE devices and hence, NB-IoT is ‘invisible’ to them. NB-IoT devices on the other hand have a limited bandwidth and thus they only see their NB-IoT resources but not the larger LTE channels. And yet, they are both operating in the same ‘channel’. In other words, LTE for broadband is one (radio-) network slice and NB-IoT is another network slice, with quite different layer 2 properties in terms of bandwidth, data rate, and channel arrangements. LTE was not designed from the start for embedding another air interface so some compromises had to be made. But there you have it, LTE + NB-IoT is network slicing in action today. Network operators can either handle the LTE and NB-IoT subscribers with the same core network or they can use two independent core networks (MME/S-GW + P-GW), i.e. two core network slices. Using different core network nodes can be useful as NB-IoT traffic can be handled differently in the core than broadband LTE data. For details see my posts on NB-IoT.

5G Network Slicing

The 5G radio network (5G-NR) is designed from the ground up to support different layer 2 configurations in a single air interface channel without making costly compromises, i.e. additional overhead like in LTE to run together with NB-IoT simultaneously. TS 38.300 states that a mobile device should support at least 8 different slices but the network could support many more. Within a slice, Quality of Service (QoS) like already used in LTE today can be used to prefer some data packets over others. Further, I take it from 38.300 that slices do not necessarily have to have different layer 2 configurations on the air interface. With these two points, it’s easy to imagine how network slicing and net neutrality could not get along too well in the future. But that’s another topic not to be dwelled on today.

VoLTE and Network Slicing?

This makes me wonder if VoLTE, or whatever operator based voice services will be called in 5G, would be a service that could be put into its own slice!? In LTE today, VoLTE establishes a default bearer that is separate from the default bearer used for Internet connectivity. In practical terms this means that VoLTE uses a different virtual network interface on a device with a different IP address from the virtual network interface used for Internet connectivity. On top, VoLTE establishes a dedicated bearer for the voice packets that receive preferential treatment in the core network, radio network and air interface to ensure a constant and low delay in fully loaded cells. In other words, these packets are preferred over the Internet packets of the same and other users of the cell. In addition, methods like automatic periodic uplink transmit opportunities can be configured for VoLTE speech packets that are sent at a constant interval to reduce the signaling overhead. All of this works well today in LTE but perhaps network slicing could be used in 5G to further separate VoLTE from Internet connectivity?

Connecting to 5G Core Network Slices

Another interesting point TS 38.300 describes is that a radio network slice is connected to a core network slice. When a device attaches to the 5G radio network the gNB has to decide to which core network ‘slice’ it has to send the signaling messages. Again let’s look at LTE today: In a combined LTE/NB-IoT access network that connects to one core network (MME/S-GW) for LTE and another core network for NB-IoT this decision is simple. If a device contacts the eNB via LTE, signalling messages are sent to the LTE core. If a device starts communicating via NB-IoT, the signalling goes to the core dedicated to NB-IoT. In 5G network slicing, things will not be as simple as several radio network slices could use the same layer 2 configuration so this can’t necessarily be used to make the decision. In its current version the specification is still a bit vague but mentions that the device can help in the decision making by sending a temporary ID or (slice selection) ‘assistance info’.


Broken down like this, the concept of network slicing suddenly becomes a lot clearer.