It is the ability to support multiple subnetworks simultaneously, with different performance characteristics, on a common physical infrastructure. A network slice can be composed of both physical and network resources and is implemented at both the data and control planes.
Quoting the IETF directly: “A network slice is programmable and has the ability to expose its capabilities.” This sometimes overlooked aspect is essential for dynamic control. It is useful to think about a network slice as exposing an API. If there is no API, then it is not a true network slice.
What Problem Does Network Slicing Solve?
It enables efficient delivery of a broad mix of services with different performance characteristics and associated SLAs on a common network. Many foresee network slicing as necessary to support 5G’s multiple services classes, particularly, ultra-reliable low latency communications (URLLC). As such, network slicing is often thought of as purely a 5G-related technology. However, it does not need to be, and in the case study, we show how it can deliver value even for fixed access consumer services.
Read more about Network Slicing
Don’t We Already Have 5G Network Slicing Today? (e.g. MPLS VPN)
An MPLS VPN is indeed an example of a network slice and has shown its value in replacing private lines. However, it is very complex and expensive to implement, which is why it is limited to fixed-access business services. In addition, while MPLS VPNs are programmable in theory, their complexity has dictated that almost all implementations are static.
5G mobile services are intermittent or on-demand by definition. If we are to apply network slicing to 5G and consumer services, we will require a much more dynamic, efficient, and scalable approach.
Why Haven’t We Implemented Dynamic and Scalable Network Slicing Yet?
It is essentially a chicken and egg problem. Stated another way, it is an “If you build it, they will come” situation. No one has built it yet, even though all of the technologies already exist.
When Apple created the iPhone, they did not need to develop any new touch screen, microprocessor, or any other technology. Their innovation was in integrating existing technologies into a package that they thought customers wanted and were willing to pay for, without customers actually asking for the new iPhone. At the same time, other companies focused on legacy phones with cost in mind, while conducting market research on what customers said they wanted, which of course was just variations of familiar experiences. It can be a similar situation for dynamic and scalable network slicing. The data plane technologies already exist in packet routers and switches, as well as optical switches. So do the SDN and control plane technologies. There must be a more profound reason why telcos are not pursuing network slicing as a means to expand service differentiation. The answer lies in economics and the inertia of existing business models. In order to maximize profitability, service delivery businesses usually develop schemes that create differentiated service offerings – at different price levels – based on differentiated demand for factors, like speed or quality of experience. Examples include package shipping, express lanes on a highway, or seats on an airplane.
Aside from expensive fixed-line business offerings, like private lines or MPLS VPNs, telcos have been reluctant to explore this model further. As a result, they are inevitably losing business to over-the-top innovators, who are not only using the power and ubiquity of Internet-based (best-effort) services, but are also starting to chip away at differentiated connectivity services. For example, AWS (Direct Connect) and Azure (ExpressRoute) already offer direct connectivity to their sites, thus bypassing public Internet access with its unguaranteed performance. Even more, AWS offers a new platform and service called Wavelength that aims at delivering ultra-low latency applications from AWS based network edges to mobile devices and end-users, positioning themselves as 5G communications suppliers. This is truly an absurd situation. To deliver Wavelength, AWS is creating a network-edge computing overlay on top of existing telco networks, which telcos are in a much better position to deliver in the first place. (For further reading, see the recent LightReading blog, “Telcos Let FAANGs into the Edge at their Peril.”)
5G may very well be the last opportunity for telcos to change their business model, from being just a supplier of raw bandwidth, to being a supplier of differentiated connectivity services. To do this, they need to be able to deliver “MPLS VPN-like services” to businesses and consumers, but dynamically, on a pay-per-use basis, and much more cheaply. For such challenges, they must implement a combination of soft and hard network slicing.
Read more about Network Slicing
What Is the Difference Between Soft and Hard Slicing?
The following brief answer will become clearer during the Case Study discussion.
Soft slices provision resources in such a way that, while the services they carry do not, on average, interfere with each other (and one service cannot receive another’s packets), they usually compete for resources, such as their position in a buffer queue, or CPU cycles. As a result, services running over soft slices can only be engineered to have an average level of performance. A relatively higher or lower level of performance is achieved by adjusting the degree of “contention” allowed for these resources. However, performance can never be guaranteed absolutely. Just about all forms of packet-processing-based services are examples of soft slicing implementations, including MPLS VPNs.
Hard slices provision resources in such a way that the services that they carry are fully isolated from other services, short of network failures. As a result, services running over hard slices can be engineered to have an absolute or guaranteed level of performance. Examples of resources used to build hard slices are TDM time slots (time isolation) and WDM optical channels (frequency isolation).