Network Slicing: The Long-Awaited SDN Killer App?
Network slicing is the ability to carve out and dedicate different sets of resources, end-to-end, from a common network, to support different types of service. It is tightly linked with 5G mobile, where its role is to magically provide networking support for 5G’s multiple service classes – headlined by high-speed mobile broadband, low-latency critical communications, and massive IoT.
As the networking requirements for these service classes and their combinations vary widely, network slicing is seen as mandatory to creating an economical 5G infrastructure. The unacceptable alternative is to create multiple supporting networks.
Once it is established that 5G necessitates network slicing, the assertion that invariably follows is that network slicing itself “will be enabled by NFV and SDN.” These are the magical components. Let’s examine their roles more closely.
Is NFV indeed a necessity? After all, it just virtualizes existing functionality, like EPC, IMS, and DPI, and does not add anything new. The argument for NFV is that its flexibility in turning virtualized functions up and down over a common set of computing resources will have a positive economic impact. This makes sense intuitively, particularly with virtualization at the massive scale envisaged in a 5G ecosystem. So while NFV may not be essential, it is highly beneficial to the 5G and network slicing business cases.
SDN - and to be clear we mean SDN-for-the-WAN - needs a deeper analysis. In the data center, SDN’s role was well-defined, separating the IP networking control plane from the packet forwarding plane. However, since SDN moved into the much more diverse and complex world of the WAN, it has taken on the much broader meaning of any software-based control of the underlying network. This has benefited marketers who have been able to jump on any new programmable feature and call it SDN, including putting lipstick on the front-ends of network management systems.
While there may be no hard definition of what constitutes “real” WAN SDN, I suggest that its principal objective is to enable network automation, and to that end, it should include the following capabilities:
- Real-time resource allocation – assigning network resources dynamically to services or connections.
- Multilayer path computation – finding the most efficient connectivity, considering the totality of L1-L3 resources, to support real-time service provisioning, dynamic shared resource restoration, and overall network optimization.
- Service and function chaining – setting the order in which functions like security, forwarding priority, and other policies should be applied to a service.
- Analytics – maintaining an accurate view on the state and utilization of the underlying network resources.
- Coverage of multivendor networks – doing all of the above in a vendor-agnostic manner, to enable best-of-breed capabilities and avoid vendor lock-in.
- Powerful northbound interfaces – enabling the automation applications.
Perhaps this is a reverse engineering argument, but it is clear that each of these capabilities has an important role to play in “carving out resources and effectively managing them in an end-to-end network”, the definition of network slicing. And this particularly true when you combine these capabilities together. So while WAN SDN has been slow to find its true footing, it appears that it will be essential to realize an economical 5G service and network ecosystem. Network slicing may, in fact, be the killer app that provides WAN SDN the focus to realize its full potential.