Network Functions Virtualisation (NFV), as everyone probably knows by now, is about de-black-boxing the telecoms infrastructure (as far as today’s technology will allow). What today is a slab of software tied to a particular proprietary computer system or network device, is to be virtualised so it can run with other de-boxed functions on standard X86 servers.

Advantages include: stunning lowering of capital costs; stunning lowering of operational costs, more fine-grained dimensioning and allocation of the available resources and - thanks to software control and a high degree of automation - greater service agility, flexibility and openness. Oh, and the promise of greater customer control over services.

So many advantages that there MUST be a catch somewhere, you might think? If there is it’s this: to get to the promised land telcos will have to reasses their approach to all the ‘r’ words. That’s reliability, resilience and ‘rithmatic (OK just 2 ‘r’ words). The proponents of NFV say its not that reliability will be any less, just that it will have to be assured, measured and offered to customers in a different way.

In effect, instead of hardening, certifying and calculating a string of 9s (99.999 per cent reliability) to each network component, the 9s are to be derived by cloud redundancy. If a virtual function goes down on one server it can be picked up by another - perhaps in a different data centre. That’s the theory.

But what about raw performance? say some interested parties. While there are a lot of interesting ‘proof of concepts’ developed, there are use cases in NFV where carrier grade performance in all its guises will still be essential. For Glenn Seiler, vice president, product management networking solutions at 'real time' specialist Wind River, a subsidiary of Intel, real-time virtualisation in many cirucmstances is going to be crucial: "You can't sacrifice performance just to have virtualisation," he tells Martyn Warwick in one of our featured interviews.

Wind River has announced a 'hardened' NFV platform for ISVs to run their more critical, deterministic applications on. "It's hardened or real-time virtualisation that will drive performance," he claims. "You still have to have 5 or 6 nines."

That’s an observation that would probably be endorsed by TM Forum chairman and CEO, Keith Willetts, who told Guy Daniels that NFV is not just another network technology. It’s completely holistic - everything changes and some big questions bubble up.

“How do you get 6 nines out of it and how do you get revenue out of it?” he asks. “There are significant opportunities and signficant threats… and… we have to rethink the dynamic of security, authentication and reliability to deliver it.

“But you also have to take a ‘good enough’ approach as opposed to [finding] a perfect solution that will last 30 years. Everything is moving so fast that you could spend longer specifying things than the lifetime of the technology,” he says.

It’s back to the conundrum of the 9s v. bare metal servers again, and whether you can really get the 9s while keeping the servers.

Hewlett-Packard’s NFV champion, Bethany Mayer, senior vice president and general manager, HP Networking, might well be spending some of her talk-time with telco executives going over this very point. She told Martyn Warwick that CSPs need to move fast if they’re to compete properly with the OTTs.

“It's very important to make this change now as, frankly, they’re going to lose the competition to the OTTs if they don’t. They have to have a change in their mindset as well. Instead of taking this new concept and applying some of the old paradigms likes NEBS-level certification or 6 9s. Instead of focussing on that they need to be very agile.

“Yes, this [NFV] is carrier grade, but are they sure they need ALL the carrier paradigm. It’s a bit of a trade-off to them that they need to think about carefully. “