|Topic:||“Radical future – Pragmatic beginnings”?|
|Author:||Dirk Trossen/ Phil Bridge|
|Title:||Principal Engineer of InterDigital & CW Virtual Networks SIG Champion/|
Senior Network Architect of EE & CW Virtual Networks SIG Champion
Principal Engineer of InterDigital & CW Virtual Networks SIG Champion
Philip Bridge has spent over three decades in the telecommunications and networking industries. He has worked in research labs, academia, start-ups, equipment vendors and service providers, in the UK and Switzerland, in technical, marketing and management roles. From 2002 he worked at Orange as a network designer and architect, and continues that role at EE since it was formed from the merger of Orange UK and T-Mobile UK. Currently he is concerned with the technical strategy for mobile data in general, and the evolution of the architecture of the mobile data core network in particular.
Phil Bridge, Senior Network Architect of EE & CW Virtual Networks SIG Champion,
Dirk Trossen has more than 15 years of experience in network architectures, wireless technology, and context-aware services. He has led numerous research projects in his corporate positions as well as in international collaborations with world-leading universities and institutions like MIT, Cambridge University, GeorgiaTech and Columbia University. Prior to joining InterDigital as a Principal Engineer, Dirk held a position as Senior Researcher at the Computer Laboratory of the University of Cambridge. He also held prior positions as Chief Researcher with BT Research and as Principal Scientist with Nokia Research. He is a research affiliate with the Advanced Network Architecture group at MIT CSAIL and one of the co-founders of TecVis LP.
In future, network capacity has to scale massively and fast to cope with a ‘data tsunami’, composed increasingly of video that aggressively expands its throughput demands to give users the highest quality experience. Networks also have to scale massively in the number of connected devices – a trend that will only accelerate as we move towards an ‘Internet of Things’.
For some time now there has been a growing perception that the way networks have been traditionally built and operated will not meet the business and operation challenges expected in the future. Network capacity has to scale massively and fast to cope with a ‘data tsunami’ composed increasingly of video that aggressively expands its throughput demands to give users the highest quality experience. Networks also have to scale massively in the number of connected devices – a trend that will only accelerate as we move towards an ‘Internet of Things’.
A consequence of all of this is that operators face increasing costs at the same time as margins are under growing competitive and regulatory pressure. It also makes networks complex and inflexible, at the same time as operators are trying to roll out services faster.
It is unlikely these underlying trends will go away any time soon. But it has become apparent to many observers that the way we build and run networks is also part of the problem. There are too many disparate, specialised components, too much hand-crafted configuration that has to marry up end to end. It takes too long to test a new component and ‘plumb it in’ to the existing infrastructure.
SDN [Ref https://www.opennetworking.org/sdn-resources/sdn-library/whitepapers] and NFV [Ref http://portal.etsi.org/nfv/nfv_white_paper.pdf] have their roots in the realisation that the IT industry has already addressed similar challenges, and that many of the resulting ideas can be adapted to networks. NFV is about decoupling network functions from the underlying hardware so they can be run as virtual machines on COTS servers in a way that has become the norm in the IT world, while SDN is about putting the network under programmatic control, coupling it with the IT world and opening the way to true automation. Each is powerful in its own right, but together they create a very powerful new paradigm that will revolutionise not only how we architect networks, but the structure of the industry itself.
The revolution is already under way and gathering pace. Almost all network established equipment vendors now have ‘virtualisation’ and ‘software control’ of their solutions on their roadmaps, if not as actual products, and many operators have strategies in place for introducing the new technologies [Ref http://www.ericsson.com/res/docs/whitepapers/wp-sdn-and-cloud.pdf and http://www.sdnzone.com/topics/software-defined-network/articles/369811-nsn-introduces-new-nfv-telco-cloud-solutions.htm]. The disruptive nature of the trend is also becoming apparent, with new vendors entering the market with solutions explicitly based on the virtual approach. Some major operators have announced plans to migrate large parts of their estate.
However, there is also a lot of challenge and uncertainty. Operators will need to develop skills in the new technologies and new commercial as well as support models will be needed. It is likely that new services and business models will emerge to support operators in building and running virtualised networks. For this reason many operators are adopting an incremental approach to introducing the new technologies, migrating individual components or functions to a virtualised/COTS model as and when they become ripe for expansion or replacement [Ref http://www.globalservices.bt.com/uk/en/insights/nfv and http://www.lightreading.com/carrier-sdn/nfv-(network-functions-virtualization)/nfv-highlight-ee-provides-reality-check/d/d-id/707096]. This delivers cost benefits without fundamentally changing the overall architecture, as well as developing an understanding of the issues involved in sourcing and supporting the new technologies.
We are also seeing virtualised, software-driven solutions already coming onto the market that go beyond simple cheaper equipment and enable new network capabilities. A prime example is so-called service chaining, where network functions can be automatically connected in a sequence specific to an individual subscriber, opening the way to mass-customisation of network services and vastly accelerated deployment times for new services. We can expect to see such solutions increasingly being deployed in the medium term.
But it is in the medium to long-term that we can expect the real revolutions. As the number of virtualised components grows and operators become confident in how to procure, deploy, support and operate the new technologies virtual networks will enter the mainstream. As the key domains in operator networks come up for refresh this will be done on a virtualised basis. Eventually a critical mass will be reached where radically new network capabilities start to emerge.
One example is the collection and analysis of data about the traffic being carried. Operational monitoring of control as well as data plane traffic is crucial, especially in mobile networks, but the cost of collecting the data and feeding it to dedicated probes for analysis is becoming disproportionate as traffic multiplies. An area of huge potential growth is the application of Big Data analytics to network traffic, but again the cost of collecting, storing and transporting the huge volumes of data is a barrier. By embedding the collection and analysis of traffic data in the same infrastructure as the virtual network components the costs should be greatly reduced, and the flexibility in which data to collect and how to analyse it greatly enhanced.
But the impact of Big Data on network operation can also be driven from applications such as online social networks. Feeding application-level data into the control plane of future networks can open up opportunities to reflect, e.g., social, relationships in the way that the network is configured. For instance, more resources in your home network could be granted to your Facebook friends while standard policies apply to the friends of your friends, with no access to your home by anybody else. Enabled through the programmability of future networks, this would bring a social face into networking (‘Social SDN’) [Ref http://www.ewsdn.eu/previous/presentations/Presentations_2013/vpun_ewsdn_panos.pdf].
Another example is network segregation. The cost benefits of having all services on a single infrastructure have always been obvious, but difficult to achieve in practice. Building a dedicated physical ‘silo’ has been the fastest way to bring a new service to market, the resulting fragmentation of the network difficult to integrate after the event. And there are sound operational reasons for keeping some services separate from others, for example voice and data services. Although a partial solution is possible by segmenting individual devices into virtual ‘slices’, and by using VPNs, the result is often very complex and difficult to manage.
Virtual networks offer not only the possibility of each service having its own instance of a particular function, but maintaining the segregation end-to-end, in effect building complete virtualised networks for each service, dimensioned and configured to reflect the particular attributes of each service, on the same, low-cost COTS infrastructure. It becomes conceivable to build a (virtual) network for voice, or M2M, or IoT if it requires attributes sufficiently specialised. And into the bargain network stability is enhanced by containing issues like signaling storms in one virtual ‘compartment’.
Another possibility is enabled by the NFV aspect of virtual networks. The IT technology employed to build virtual networks rests on the concept of dynamic ‘orchestration’ of the processor and memory/storage resources assigned to each function. Virtual machines can be automatically created and released as load on a particular functions rises and falls. Taking this idea one step further, resources can be allocated and switched between functions depending on the operator’s policies. Virtual networks could be built as a service for 3rd parties, within a ‘container’ specifying how much total resource is allowed. And if dynamic policies are informed by Big Data analytics, itself enabled by virtual networks, the responsiveness and efficiency of networks increases further.
But perhaps the most profound long-term change will come in the way networks are procured and operated. It is not a great leap to foresee the emergence of infrastructure-as-a-Service (IaaS) providers, who own and operate the physical assets – the fibres, the facilities with power, space and cooling, the standardised server and switching environments. The synergies and cost savings of this model are probably too compelling for it not to happen. What is less clear is which entities will take on the role of building the virtual networks on top of this infrastructure. Will networks be offered as-a-Service (NaaS)? Will the 3rd parties that have APIs into today’s networks be able to build their own virtual network via a NaaS? Such prospect would open up opportunities to provide applications in the form of a product-as-a-service (PaaS) that encompasses the entirety of the application logic and the virtual network that realises its distributed functioning in a controlled and quality-ensured manner. Such joint orchestration of application and network resources would ultimately ensure that the quality of experience can be tailored to the specific application’s need while segregating the application product from possible negative impact caused by others in the physical infrastructure.
One thing is certain: virtual networks are about more than incremental CAPEX and OPEX savings. Over the next five years or so they will trigger a radical re-shaping of the network industry. What remains uncertain is exactly what this future will look like.