So, we've talked a little bit about 5G and some of the use cases in 5G, and we've implied that there're going to be impacts into the network. We're going to dig down a little bit into that now, in area around network slicing. So, network slicing is the word that we're using to talk about the allocation of the resources in that network, and to discriminate them into certain functional categories based on the needs of those large number of devices that are going to be in the network. So, for example, in this slide we talk about smart meters and the device massively scalable, millions of them or hundreds of millions of them in many cases, they're signaling plane usage from a control plane perspective is relatively low. We're looking at octets per hour thing. Their data plane throughput is almost non-existent, very insignificant for that point, and absolutely no mobility when we think about these devices, and they also have a great tolerance for latency. So, we have a very large number of these devices coming into the network and sending traffic into them, the burden on that network is not significant, and at the same time it's very tolerant of other uses of the network, for a non-consumer video. So again, we're thinking about center monitoring from a video standpoint, relatively moderate to high in a number of those devices. Again, relatively low control plane throughput, the data plane throughput in the low to moderate range, and again there's no mobility aspect associated with it, and because of the nature of video, its tolerance for latency is actually is going to be relatively low, as opposed to the smart meters where we say connected cars. Depending on the exact type of applications, we've talked about a couple different things when we're speaking about connected cars, but obviously, as that begin to scale up on a relatively high number of those devices in millions, tens of millions in many cases as we move forward in time, their control plane load is going to be high, it's going to have a rather significant throughput, as is their data plane throughput, so both very high demands on that, and obviously from mobility standpoint, this is one of the prime candidates that we talk, about when we think about. The latency, there may be some tolerance depending on the functionality that goes into place but it probably fits somewhere between what we'd expect to see on that video scale and certainly the smart meters. Then finally, we've got the traditional user, they are the smartphone users, in common parlance the UE, the user endpoint. Tens of millions of those devices so relatively high number, their control plane traffic is a little bit higher than you might have expected to see in some of that consumer video area, but falls into that moderate range. Data throughput, while today we may think of it on the 4G network as being very high, because that the primary consumer of that traffic in the 5G network that's going to fall into that moderate range from a mobility standpoint. Again, these users maybe on the Wi-Fi as much as they're on the macro network. But when they're on the macro network, their mobility is going to be higher than it is when they're on one of the unlicensed like Wi-Fi type networks, and their tolerance for latency and actually going to be driven by the application itself. If they're doing something like a video-voice type of call, their tolerance for latency is going to be significantly less than if they're streaming some sort of video, we've got a large capacity for buffering it if it's a non-real-time type video. So, depending on the application that we see as we slice things across here, and that's where this comes from, is we slice things across here, we may have different burdens on the network itself. So, pictorially visualizing this, then from an autonomous driving vehicle to some type of smart meter device, to some UE type device working from bottom to top here. Those will all come into the same range, in many cases, on the macro network presumably, the 5G network then it fall back to 4G network and then the range is going to transfer that device and maybe identify it, because it can do that, it can go based on some unique identifier associated with a chip that's in that device. There's some chip, for example, it may associate with a particular APN, an application identifier if you will, they could say, "Hey, this looks like a UE or this looks like an autonomous car, because it knows the identifier. Then at the core of the network, we can treat that differently even as that traffic goes all the way into the cloud." So, we see that representation pictorially than in the oval diagrams here, where smartphones may get a different slice than the smart city or some vehicle to vehicle type of device. In particular, it's not just the slice itself but the elements that were allocated or the volume of the compute resources or the volume of the network resources that are allocated to that slice can be determined based on the collection of those devices. Again, when we think about this, you've got to remember that there are lots of these devices connected to lots of brands and there're lots of brands connected to lots of cores into the network and we want to be able to manage those elements as a ubiquitously as we go through the slicing of the network into a site. So for example, the RANs themselves depending on your geography, you may find that those radio access networks or in a distribution of every 10 to 15 kilometers or less, the core networks, if you've got a very large geography, for example, like the continent of Australia, the country of Australia, very large geography relatively lighter population standpoint maybe able to be covered by a couple dozen of the core network elements, you take that same geographical size. For example, in the continental United States to 48 states, and you may find that you've got an order of magnitude more of those core elements because the density of the population in there. So again, it's not a one location that comes in, is that these networks are significantly distributed geographically and based on that workload that's in there. So, we want all those elements, then as we begin to look at this transformation to be able to support this concept of that network slicing, so that again we're not over-provisioning that network and we get the right amount of resources for that burden, for that workload, for that user that consumer of those resources inside the network.
