Did you know that “cloudify” is now an English verb? I didn’t.
“Cloudify” … and “cloudification” were buzz phrases in Madrid this week at Informa’s Network Virtualization Europe conference. In fact, as I learnt, “cloudification” is the half-step between turning legacy network elements into virtual machines. Expect a new entry in the Oxford English Dictionary any day now! But apart from the much-appreciated grammar lesson, what else happened? With an NFV gathering now occurring somewhere in the world every five minutes, are the only people making money from NFV the conference organisers?
As a marketing guy, I approached this event with two simple questions:
- What’s the killer use case driving virtualisation?
- What’s getting in the way of virtualisation?
Or, even more bluntly, what exactly are we trying to do - what's stopping us?
I patiently ignored all those architecture diagrams that frankly send me to sleep, and tried to stick to my two “dumb questions”.
Here’s what I heard.
1. What’s the killer use case for NFV?
The best thing since sliced bread? Network slicing of course has been touted as the secret sauce that virtualisation with 5G brings us. It enables operators to dynamically allocate virtual network resources and functionality to whatever use case demands it, moment by moment, with a guaranteed Quality of Service. Basically, you create as many virtual networks as you like on top of a common physical infrastructure.
This dynamic partitioning of a network was frequently listed by speakers as the MAJOR use case. (In fact, it isn’t a use case, no consumer I know wants to slice a network in two, but it is certainly a mechanism for delivering use cases). And it was a big conversation topic - virtualised network slices that offer ultra-low latency or huge capacity or high reliability etc, depending on the market.
One slice to rule them all!
The ever-insightful Sue Rudd of Strategy Analytics brought an interesting angle to this. She says instead of considering a “slice” vertically eg for thermostats, or smoke sensors, or security cameras, or remote live university courses, or autonomous cars etc, we should swing the various segments through 90 degrees into horizontal slices based on their Class of Service (CoS) requirements. A Horizontal network slice is essentially based on a bunch of CoS parameters such as latency, jitter, bandwidth, throughput requirements etc that cuts across many vertical applications.
Sue believes that by aggregating ‘nichey’ verticals into network slices that share similar CoS requirements, Communications Service Providers (CSPs) can “flip” poor vertical market business cases into profitable horizontal ones. Big gains can then be made by CSPs who combine similar use-cases and direct them to the same shared pool of virtual resources. So, for example, applications that all require similar latency or similar compute or similar bandwidth requirements are all taken care of by the same network slice. That requires a powerful classification mechanism so that incoming service flows are forwarded to the right slice accurately and instantaneously as they arrive. Real time classification and assignment of service flows could be an opportunity for an IP Traffic Manager perhaps?
An edge "too far"
To be sure, the key enabler for many use cases, again endorsed by speaker after speaker was (unsurprisingly) edge computing. A lot of traffic can and must be offloaded or turned around at the edge. A representative from Intel described a trial of facial recognition they undertook with Alibaba. The performance requirement was 10 faces per second which clearly can only be handled at the edge.
A speaker from BT noted “the edge” sounds straightforward in principle but reality is different. “Where exactly is the edge?”, she asked after describing BT trials. A car to traffic-light message that does not need to go back to the core is one scenario, but a stadium full of TV cameras streaming Terabytes of HD footage to a mobile van is different. So “the edge” will vary from a private site, to a cell, to elsewhere. Too far from the edge means losing out on performance, but too close to the edge means losing out on economies of scale.
2. So what’s stopping progress?
Sadly, nothing new.
A speaker from Telstra was keen to bring the industry back down to earth. Legacy silos are so entrenched both technologically and organisationally they are proving harder to break-down than anyone imagined. No more so than in network management. Traditionally every time a new supplier’s technology gets added, a new network management system is also added. Eventually it becomes impossible to remove a supplier due to the “domino effect”. Even the NFV standards are only adding to the complexity in Telstra’s opinion. Vendors who walk in the door boldly proclaiming they can provide end to end orchestration, the ultimate MANO, cannot be taken seriously. As the speaker said, “I already have 22 vendors who promised to do that … so what exactly do you mean?”.
Another operator I won’t name went even further: “50% of our legacy equipment is rarely used or never used, but taking it out is too hard!”.
“This is not about mapping elements one to one into a virtualised space”, concluded a speaker from Telefonica. “It’s too easy to make this a technology discussion – it needs to be a process discussion … it’s much more about the organisation than we previously thought”.
And one bonus question that is always worth asking …
3. Who’s making money from NFV (I mean apart from conference organisers)?
Several speakers were asked the question: “how much money is virtualisation actually saving you?”. The replies varied from slightly cagey to refreshingly honest: “ … significant savings on the CAPEX side, but we’re still working to understand savings on OPEX” said Telefonica. Real OPEX savings are a result of process change and automation - that takes a long time to implement.
There was an acceptance among several operators that, to compete with the Microsoft Azure, AWS etc cloud services, CSPs need to be perceived more as cloud service providers and less as telcos. 3 Austria honestly admitted “we are not ready yet, we are not fast enough to deliver what we need”.
A speaker from Intel noted, “with a virtualised network, there’s no reason why telcos cannot be cloud providers, but the key to profit has to be enterprise applications. Relying on SIM card subscriptions won't cut it”.
That sense of déjà vu
One panel member noted ruefully, it’s not that different to the IMS discussion years ago. Everyone agreed it made sense but: “no one use case justified the investment. You must look at it as an operator wide investment - driven from top down and end to end. Real cost savings are not made by virtualising pieces of network. They will be made by new ways of building a network”.
Opportunities to change the network on this scale only come once in ten years. Let’s not screw it up. Before we virtualise we must rethink the network design and virtualise a simpler approach. As a representative of NetNumber concluded, “If you virtualise chaos you still have chaos”.
... Now there’s a thought.
Openwave Mobility was awarded a Finalist position for Most Innovative NFV Solution, an independently judged competitive award that took place during the conference. The award was received for Stratum - a common data layer solution.