The past two years have seen the hype around network functions virtualisation (NFV) reach deafening levels. What’s the fuss?
Here’s a clue: ‘To run a data centre, the web-scale guys use one IT person to run 10,000 servers. In our space the average is one IT guy to 10 servers.’
This comes from Dor Skuler, who runs CloudBand, Alcatel-Lucent’s attempt to crowdsource its way into leading the suppliers’ response to network functions virtualisation (NFV).
His customers are way ahead of him. BT started looking at the costs of running big IP networks and services about three years ago, and how they were likely to change over the next few years.
Shocked, it started experimenting by emulating appliances such as switches, routers, firewalls, load balancers, network address translators etc as virtual machines on generic server hardware. Result? There was no meaningful difference in performance compared to proprietary systems, but at a small fraction of the cost.
BT realised that, even as a top 20 telco by revenue, it had limited bargaining power against the likes of Ericsson and Huawei and Cisco. It sought a big brother in the form of Deutsche Telekom, and boom! we were off the races. A 2012 conference to discuss BT’s findings and what to do about them attracted little more than 100 interested parties; a year later more than 300 went.
BT’s White Paper on NFV became the basis for an ETSI (European Telecommunications Standards Institute) working group to develop coherent standards for virtual network functions (VNFs).
Suppliers, already reeling from the implications of the parallel and related US-based software defined networking (SDN) initiative, which targets DCs (data centres), struggled to understand the implications for the networks outside the DC.
Time is not on their side because their customers are bleeding. A report late last year from the Telco 2.0 Initiative suggested that telcos could lose up to $172bn from core revenues in five years if they don’t make dramatic improvements to their voice and messaging strategies.
Even if they fight back using the latest technologies like WebRTC and VoLTE, at best they can preserve only $80bn. So that’s $92bn that’s going to the ‘web-scale’ guys like Google, Amazon, Skype, Akamai, in other words, the OTT players, says Telco 2.0.
The ETSI group has picked up this need for speed. It gave itself just two years to deliver workable standards that address what the conferences agree are the main pain points, namely:
• Better capital efficiency than that offered by dedicated hardware, using commercial off the shelf (COTS) hardware and virtualised servers and functions
• Better flexibility in assigning functions to hardware
• Faster innovation because it’s all software now
• Less operational complexity because of common automation and operating procedures
• Lower energy bills from migrating workloads and powering down unused hardware
• Standard and open interfaces between functions that are decoupled from hardware but still manageable.
The ETSI group published its Group Specification for the NFV architectural framework in October last year. This spelled out the ‘what’ of NFV, but not the ‘how’. It also indicated areas for further research. These include: better virtualisation layering and support for NFV infrastructures; a finer-grained decomposition of software-based functions to speed them up; better ways to manage and orchestrate VNFs; performance and scalability; reliability; and security.
That’s a lot to do in very little time. It seems unlikely to deliver the 1,000-fold improvement in labour productivity the industry needs to get on terms with the OTT players, at least in the short term.
But the recent decision in US courts to allow operators, especially MNOs, more flexibility in pricing quality of service (which may be appealed), suggests it might not have to.
What are SDN and NFV?
Virtualisation of compute and storage functions have led to huge productivity gains in the data centre. Companies now hope to achieve similar gains from virtualising mobile network operations.
There are two related but parallel initiatives, both driven by user communities: software defined networking (SDN) and network functions virtualisation (NFV).
SDN is driven by web-scale users such as Google and Rackspace, while NFV is driven by telcos such as Deutsche Telecom and BT. Many firms are members of both, and scores of vendors have joined too. Standards-making bodies are the Open Networking Foundation for SDN and the European Telecommunications Standards Institute (ETSI) for NFV.
The ETSI working group aims to write standards that will allow software to emulate in virtual machines running on commodity servers functions now done by proprietary ‘appliances’ such as deep packet inspection, network address translation, session border control, even routing and switching.
SDN advocates argue that by separating the content of messages (the data) from the addressing information needed to send it around the network they are able to find more efficient ways to do things, particularly in virtualised environments.
Key to this is a new protocol to manage the exchange of information between different real and virtual devices.
The ONF has published a standard called the OpenFlow protocol to govern the flow of information between a controller, which knows what is on the network and the relationships between them, and an OpenFlow switch, which moves the content between the network elements.
The ONF puts it thus: ‘In the SDN architecture, the control and data planes are decoupled, network intelligence and state are logically centralised, and the underlying network infrastructure is abstracted from the applications.
As a result, enterprises and carriers gain unprecedented programmability, automation, and network control, enabling them to build highly scalable, flexible networks that readily adapt to changing business needs.’
Portugal Telecom turns to NFV and SDN for its future
Portugal Telecom is putting its faith in advances in network functions virtualisation (NFV) and software defined networking (SDN) as it converts from a telco to an IT services provider, a move critical to its survival, it believes.
PT is Portugal’s fixed line incumbent and is also its leading mobile network operator. Regulators have just approved its merger with Oi, Brazil’s biggest telco, making it a global firm with about 100 million customers among 260 million people.
‘50% of our business today is based on data. We’re not just a telecommunications company, but also an entertainment one, which sells television. We committed to fibre, to 4G coverage and this is a process of technological investment that ends with the data centre,’ CEO Zeinal Bava said at the recent opening of PT’s new DC in Covilhã, 280km north-east of the capital Lisbon.
Once finished, Covilhã will be the biggest DC in Europe and the sixth biggest in the world. It is crucial to PT’s strategy of moving from selling ‘speeds and feeds’ to selling services such as connectivity and entertainment globally. And it will run on NFV and SDN.
PT’s technical director for networking and engineering Salas Pires says he is waiting for the industry to standardise its approach to NFV and SDN. He believes SDN will have commercial application first in data centres. ‘After that, the virtualisation of the functions in the network will go to the backbone (network), to the IP core, and also to the radio access network.
‘The next 4G networks will be the first to be virtualised, with functions more and more centralised on virtual platforms, and in some way have the (network) intelligence concentrated, but more and more distributed with more flexibility to introduce new functions and new services on the net. This doesn’t exist today, but it will within the next two years.’
Pires says the lack of standards is a problem if CIOs want to integrate SDN into data centres today. ‘OpenFlow is a standard, but the implementation by Alcatel-Lucent is not the same as the one implemented by Cisco. As a telco, we feel a little bit constrained, because normally when I buy a 4G platform it’s compatible whether it’s Nokia or Huawei or whatever.‘We could get first mover advantage by simply making a decision, but getting it wrong would be very costly, so we have to be careful. I believe it will only be in 2015 or 2016 that we will be able to do in the IP core network what we can do today in the data centre.’