Network slicing is mentioned in nearly every industry presentation on 5G, but typically it is vaguely defined. What exactly is network slicing, and how does it differ from all the Quality of Service (QoS) initiatives that have been around for years.
Network slicing is a specific form of virtualization that allows multiple logical networks to run on top of a shared physical network infrastructure. The key benefit of the network slicing concept is that it provides an end-to-end virtual network encompassing not just networking but compute and storage functions too. The objective is to allow a physical mobile network operator to partition its network resources to allow for very different users, so-called tenants, to multiplex over a single physical infrastructure. The most commonly cited example in 5G discussions is sharing of a given physical network to simultaneously run Internet of Things (IoT), Mobile Broadband (MBB), and very low-latency (e.g. vehicular communications) applications. These applications obviously have very different transmission characteristics. For example, IoT will typically have a very large number of devices, but each device may have very low throughput. MBB has nearly the opposite properties since it will have a much smaller number of devices, but each one will be transmitting or receiving very high bandwidth content. The intent of network slicing is to be able to partition the physical network at an end-to-end level to allow optimum grouping of traffic, isolation from other tenants, and configuring of resources at a macro level.
Why can’t QoS and related techniques do the job?
The natural question that often arises during technical discussions about network slicing is why can’t existing internet techniques handle the task? For example, the extensively deployed Quality of Service (QoS) architecture called Differentiated Services (DiffServ) is meant to classify and manage different types of IP traffic (e.g. voice, video, text) flowing over a given network. There are also other well-known techniques like Virtual Private Network (VPN) which separate and isolate traffic across the Internet using techniques like IP tunneling. Also, more recent approaches like Network Function Virtualization (NFV) are meant to virtualize parts of mobile networks. Each of these features obviously has some overlap with the functionality of 5G network slicing. So why do we need to introduce network slicing as another approach? The answer is two-fold. One reason is based on technical issues, and the other, more important, reason is business driven.
Business drivers for network slicing
Network slicing in 5G is expected to open lucrative new business opportunities for mobile operators and other newer entrants. For example, a mobile operator will be able to split its physical network resources into multiple logical slices and lease these slices out to interested parties. An electrical utility may want to take a long-term lease of a network slice for connectivity of its smart grid composed of sensors, meters, and controllers and optimize that slice for IoT devices. Alternatively, a concert promoter may want to take a short-term lease of a network slice for a week-long musical festival and optimize that slice for streaming HD music and VoIP connectivity.
[ To comment on this story, visit Computerworld’s Facebook page. ]
Network slicing will also allow new business models to evolve in the mobile market. For example, existing cloud and data center providers, such as Amazon and Google, may find through network slicing a new way to play in the mobile network space. This would also allow existing mobile operators (e.g. AT&T, Orange) to lease network slices and concentrate mainly on their core strengths which include delivering high-quality networking experiences. This is quite different from today, as currently infrastructure cannot be granularly configured and optimized for different MVNOs (tenants). Current MVNOs have purely business and billing relationships with the network owner and run their MVNO service over a vanilla network with branding overlays. Network slicing will allow the network owner to customize the compute, storage and networking functions of the infrastructure for a given Virtual Network Operator’s traffic characteristics.
Technical drivers for network slicing
The key differentiator of the network slicing approach is that it provides a holistic end-to-end virtual network for a given tenant. No existing QoS-based solution can offer anything like this. For example, DiffServ, which is the most widely deployed QoS solution, can discriminate VoIP traffic from other types of traffic such as HD video and web browsing. However, DiffServ cannot discriminate and differentially treat the same type of traffic (e.g. VoIP traffic) coming from different tenants.
Also, DiffServ does not have the ability to perform traffic isolation at all. For example, IoT traffic from a health monitoring network (e.g. connecting hospitals and outpatients) typically have strict privacy and security requirements including where the data can be stored and who can access it. This cannot be accomplished by DiffServ as it does not have any features dealing with the compute and storage aspects of the network. All these identified shortfalls of DiffServ will be handled by the features being developed for network slicing.
What is happening in the standards?
Due to the exciting business opportunities that network slicing will provide, there are a lot of related technical efforts going on in various standards bodies. 3GPP, for example, is specifying the overall Operations and Management (OAM) framework for how the owner of the physical network will manage the slices as part of its virtualized NFV network. 3GPP is also specifying the signaling and procedures required between network components and the UE to assign a device to a given slice.
The Internet Engineering Task Force (IETF) has started to examine how the underlying IP network will implement network slices. As previously mentioned, there are already existing protocols like DiffServ and VPN which support network slicing like features. The IETF is still debating which protocols, existing or new, will be developed for network slicing. However, there is an early consensus that a lot of the required network slicing functionality can be accomplished by a judicious mix of OAM and traffic engineering. The new OAM features will allow definition, creation, deletion, etc. of slices. These slices may extend between administrative boundaries, meaning that the slices may span multiple operators or countries. The new traffic engineering features being considered include possible updates to underlying network substrates like MPLS, SDN, etc. to allow control signaling for network slicing purposes.
Network slicing is very different from QoS because it will enable end-to-end virtual networks encompassing compute, storage and networking functions. Existing QoS approaches are all point solutions which offer a subset of functionality at best when compared to network slicing. Network slicing is generating a lot of excitement in 5G discussions mainly because it will open many new business opportunities. Network slicing is part of the general trend to make network services more virtual and thus benefit from the lower costs and increased innovation that the IT industry reaped with the move to the cloud and Everything-as-a-Service. It is not likely that network slicing will require revolutionary new technical standards. Instead, the technical changes needed will be applied across a myriad of technologies and standards all focused in the key areas of improved network intelligence, system integration, OAM and traffic engineering.