SDN World is a brand new conference focusing on the future of network management and took place in Barcelona, June 2013. The conference was well attended by representatives of telcos, carriers, mobile network operators and network equipment vendors.
Despite being billed as an SDN conference, the focus was actually on two approaches: Software-defined networking (SDN) and Network function virtualisation (NFV). While there were some claims that there exists confusion surrounding the two terms, all speakers seemed to agree on the following broad definitions:
- SDN – separation of the control plane from the data plane, allowing network devices to be programmed using a standard interface (e.g. OpenFlow)
- NFV – virtualisation of network functions, in effect throwing away the specialised hardware boxes and moving the ‘functions’ (e.g. routing, load balancing) into software running on standard hardware
SDN and the OpenFlow standard are overseen by the Open Networking Foundation, while NFV is the remit of an ETSI Special Interest Group. The two approaches are not interdependent. In fact, NFV without SDN is possible but difficult, so the two are complementary and often discussed together.
What do SDN/NFV mean for business?
There was widespread recognition that SDN/NFV is the next logical step in network management, analogous to the innovation in server architecture layering, standardisation and subsequent virtualisation of components. However, the communications service provider industry has been typically slow to adapt and while those attending the conference are excited about the change, they are keenly aware of the risks involved. The desire is for a slow and evolutionary deployment of SDN, while the effects on business become clearer. Given that it has taken ten years to adopt server virtualisation in the data centre, predictions are that it will be another ten years hence before we see mature deployment of SDN/NFV.
The proposed benefits of SDN and NFV are clear and were mentioned by almost every speaker. They are:
- reduced capital expenditure on the infrastructure (through lower cost of ownership through industry standard servers/storage/switches)
- reduced operational expenditure (through simplification of architecture, automated/remote deployment and management)
- reduced time to market (by the ability to innovate and deploy solutions at the speed of the software lifecycle)
- new sources of revenue (by more easily differentiating services for different customers).
Of these benefits, it was the identification of new sources of revenue that caused the most discussion. It was claimed that the savings through reduced capex/opex would barely justify investment in SDN/NFV (although this was disputed from some corners), while reduced time to market only brings revenue sooner. Indeed, overprovisioning the network may still prove cheaper than complex QoS management in the medium to long term. It was agreed that there exist many opportunities to monetise SDN/NFV, such as multi-tenancy services and QoS differentiation, but many commercial operators were either unaware of or (more likely) unwilling to reveal any concrete proposals.
There is a hope that SDN/NFV would allow infrastructure providers to compete with ‘over the top’ (OTT) providers, such as Netflix, Skype and Google, which are currently making good business providing services to consumers uninterested in the network (and without the investment in it). Carriers should investigate revenue sharing models with OTTs. There was also much focus on micro-transactions (e.g. paying for a temporary boost to the quality of service, or QoS) as being more palatable to consumers than an increase in standing charges.
It is clear that the vendor market will soon be forced to move from a hardware to a software business, resulting in companies that develop and (remotely) deploy and manage network applications rather than appliances. However, this begs the question of where operators source the commodity equipment on which to run these applications and who is at fault when the service fails.
For SDN/NFV to gain large-scale adoption, people must change the way they do business. The technology is largely ready to be deployed, but the wider socio-economic impact will lag behind the changes it brings. There is a feeling that the ecosystem for SDN/NFV needs to be prepared before we will see a proliferation of network application developers and attendant innovation – a kind of “build it and they will come” approach. Critical to this is the adoption of open source. Proprietary devices and interfaces are not seen as being viable in the future marketplace. This again raises the question of how new and existing businesses will make value from the distribution and support of open-source software. Many people are looking to Open Daylight to provide a common basis for SDN.
What are the technological hot topics?
The SDN architecture talks about the southbound and northbound interfaces. The southbound interface is used to program the network switches, among which OpenFlow is the current major contender. The northbound interface is the API exposed by the controller to application developers, which is currently not standardised and differs according to the chosen controller. There was recognition that these interfaces are not sufficient and that the next major problem will come from the ‘east/west’ interface that allows controllers to talk to each other. This encompasses both intra- and inter-domain communications, although the latter is at the very farthest extent of many people’s current concerns, as they have yet to solve the problem of SDN within their own network.
There are some proposals for the scope and responsibilities of a centralised ‘controller of controllers’, ‘super-controller’ or ‘orchestrator’ that handles the automation of business-level decisions, while controllers focus on programming the switches, e.g. as per this, or this. In the SDN architecture, the orchestrator component is essentially another application using the northbound interface.
Whether there is a centralised orchestrator or a distributed network of controllers, there is a recognition of a semantic gap between management applications and network devices. There is a need for an abstraction layer that allows models of services to be mapped to models of devices, so that service life-cycle operations can be transformed into an automated, atomic and verifiable process of configuring network devices. NetYCE, and tail-f’s YANG modelling as part of Deutsche Telekom’s Terastream trial claim to bridge this semantic gap.
QoS management for end-users is generally thought to be the hardest problem to solve using SDN, because it requires control of the end-to-end network. It is difficult for many network providers to even conceive of the inter-domain orchestration required for end-to-end QoS management. A world with SDN everywhere is hard to imagine, while a world in which all parties manage each other’s traffic according to standard or agreed terms is considered many years (if not decades) away. In any case, QoS guarantees are unlikely to be about upholding minima and will mostly involve lifting a cap in exchange for payment.
SDN/NFV does not yet deal with QoS, but the two principal methods envisaged for implementing QoS are:
- priority scheduling of traffic flows, e.g. using queues and based on half a dozen traffic classes
- dynamic routing of traffic flows, e.g. selectively routing through optional network functions such as deep packet inspection, protocol optimisation, advert stream insertion.
What do regulatory authorities make of SDN?
The telecommunications industry is highly regulated. There was a keen sense that the application of SDN/NFV is a threat to net neutrality, as it allows traffic more easily to be prioritised, potentially allowing operators to squeeze OTT traffic in favour of their own (or subsidiaries’) services, e.g. like this. However, many network operators feel it is necessary to explore the opportunities for new revenue that are brought by novel collaborations and business models enabled through SDN. Regulation typically lags behind technological change and this innovation is no different.
In some cases, it is recognised that regulation hinders innovation, as it requires operators to provide costly but socially necessary services (e.g. rural communications) that do not allow them to compete with other more agile providers, such as OTTs, which may not have such obligations.
Regulation differs from country to country. For example, net neutrality laws in the Netherlands are very strict and would seem not to allow for any optimisation of traffic flows. In contrast, other countries, such as Saudi Arabia and South Korea, are blocking OTT providers (e.g. Viber, WhatsApp) that are competing with network operators’ own services, or raising the price for their use, thereby reducing consumer choice.
