NFV (Network functions virtualization)

5 Challenges for NFV in the 4G Core Network

In my previous column, I wrote about the opportunities Network Functions Virtualization (NFV) offers mobile operators, especially in the core network. (See 5 Opportunities for NFV in the 4G Core Network.)

In this piece, I discuss some of the reasons why NFV is not yet ready for prime time. I'm paraphrasing, of course, but in my work with mobile operator technology executives, I know these are the sentiments widely held by the people responsible for designing and operating 4G core networks.

In no particular order:

1. Immature technology and product availability: The performance of virtualized network functions (VNFs) is not yet equivalent to today's built-for-purpose products, or even sufficiently close to it. Reliability is also an unknown quantity and must be approached with a fundamentally different mindset. The even bigger challenge is that most operators are not close to being able to run VNFs in a "cloud" environment. This is a solvable problem, it's just too early.

2. Operational processes: You can have the most progressive CTO team in the world, but unless the proposed solution is accepted by the team responsible for running that part of the network, and making sure it hits the reliability and performance targets, it won't make much difference. Operators are used to buying "vertically assured" products from vendors that will stand by the box if something goes wrong. Changing this mindset at an abstract level is one thing. At an operational level it is quite another.

3. Investment cycle: Progressive operators have recently installed a new evolved packet core (EPC), and related core network elements, to support the introduction of LTE. Many have also recently closed tenders for IMS/VoLTE. It is out of the question to revisit these investments for at least the next three years. Any virtualized solution must be positioned with this market context very much in mind.

4. Adding new NFV platforms increases opex: Using a virtual EPC as part of a "cap and grow" strategy alongside the existing core network is commonly proposed. But while a parallel core network may make sense in terms of being able to experiment with NFV for less critical services, such as M2M, it also adds costs through duplication. This especially impacts operating costs and results in the exact opposite of what NFV promised at the outset -- reduced opex.

5. Need to identify use-cases: If the mobile network operator is basically in the business of selling Internet access, there are limited opportunities to do clever things in the services domain. So while NFV promises greater software-configurable service agility, without clear use-cases and revenue models, it's hard for mobile operators to make the business case. Whisper it, but in this respect NFV is a little like IMS.

The operators I speak with say none of these reasons alone, or even in combination, mean NFV won't happen. The challenge is to identify NFV use-cases that navigate these challenges without throwing decades of accumulated learning out the window.

And, of course, there is also a risk of being over-cautious and hamstrung by legacy processes.

Never a simple life, huh?

— Gabriel Brown, Senior Analyst, Heavy Reading

Interested in learning more on this topic? Then come to Ethernet & SDN Expo, a Light Reading Live event that takes place Oct. 2-3, 2013 at the Javits Center in New York City. Co-located with Interop, Light Reading's Ethernet & SDN Expo will focus on how the convergence of Carrier Ethernet 2.0 with emerging carrier software-defined networking (SDN) and network functions virtualization technologies could change the whole telecom landscape for service providers. For more information, or to register, click here.

Dredgie 9/23/2013 | 6:33:59 PM
Re: Spot on... Agreed. Plus - Even w/o traffic segmentation, the coexistence of virtualized and non-virtualized network functions is something the ETSI ISG highlighted when determining (and documenting) the initial 9 use cases. For the mobile core network this includes the question of whether new operation support systems are needed or existing operation support systems need to be enhanced to support a coexisting vEPC and non-vEPC. Or even if a failover mechanism to non-virtualized NF's are required.
AntonioRodriguez 9/19/2013 | 5:46:51 AM
Re: Spot on... Thanks Gabriel.

Indeed a parallel Virtual EPC coexisting with the classic EPC is what most operators seem to be thinking on right now. It makes both technical and business sense for technologies like M2M or use cases like distributed congestion reduction in specific areas, and that is what I have seen in the trials in the field so far.

The most common approach is using RAN with MOCN or RAN sharing activated for distributing the traffic to the different core networks as required. This in example makes the M2M traffic profitable again thanks to the reduced cost of the Virtual EPC.

Best regards,

Gabriel Brown 9/19/2013 | 5:37:37 AM
Re: Spot on... Thanks Antonio. I'll have a read of your blog. I agree that for now in terms of "pure network functions" the focus is on PoCs and trials.

What do you think about the idea of a Virtual EPC deployed in parallel to the classic EPC to support specific use-cases -- for example a virtualised M2M core or a progressive enterprise service of somekind?
Gabriel Brown 9/19/2013 | 5:34:15 AM
Re: NFV and SDN are ready now for certain use cases Hi Jeff. Thanks for your comment. I agree with a phased approach. This migration path is an area we're looking at closely.

Operators have identified services layer functions, and video in particular, as most attractive for virtualization. So that aligns with your deployment. In a mobile core context these "off-3GPP" services deployed on the Gi/SGi do look attractive candidates.

There's a school thought that the services layer is not equivalent to "pure networking functions" in terms of performance requirements. In your example you're talking about using functions on a P-GW of edge router to route flows to your video processing functions. The network elements themselves are not yet virtualized.

Mavenir is interesting because it is a virtualized "pure" network function – albeit it is the application server rather than the control layer and media layers of IMS (which will come in time).

We actually see a lot of operators with some virtualized services and/or virtualized control-plane functions. So far these tend to be "point solutions" to address a particular need or opportunity. You'd be hard pressed to call this as Cloud NFV. We've been using the phrase "virtualized but not cloudified" to describe this.

AntonioRodriguez 9/18/2013 | 5:27:04 AM
Spot on... This article is spot on reflecting the reasons why the operators are not still fully embracing the NFV. Although this is not stopping them from making trials already.

I have written a piece about the topic here:


with a hands-on review on NFV solutions for the EPC.

MtnViewJeff 9/17/2013 | 6:03:32 PM
NFV and SDN are ready now for certain use cases

Gabriel, many of your points are spot on but there are aspects of NFV that are deploying this year, and smart Tier 1 operators are all planning for SDN and NFV

At Skyfire, we're live in a Tier 1 operator with cloud-based video and data optimization, running in the carrier data center.  A software defined instruction set in a router at the Packet Core can filter off large objects for optimization, based on policy and QoE real time metrics.  This is not 3 years off, but now.  Telenor just awarded a sole vendor contract to ALU and Skyfire for 12 countries for this state-of-the-art NFV approach, with deployment underway now.

You're right that no operator is ready to move to a virtualized network overnight in one giant leap.  But certain functions, like Mavenir for VOLTE or Skyfire for video and media optimization, can run without ripping out the existing network, just adding new functionality.  Data centers are perfect for compute-intensive functions like video transcoding, especially where demand is hard to predict and more or less compute can be allocated each time period of the day based on demand.

Smart Mobile operators will likely follow a 3-step progression to NFV. It's difficult to leap directly to virtualizing the entire network, including EPC and Gi LAN and all of its functions, so operators will take a step-by-step, lower risk approach.   

For step 1, service providers can prove the viability of individual virtualized functions starting now, such as the traditional in-network function of data and media optimization and traffic shaping. Video and multimedia optimization in the cloud is possible when intelligent routing and steering partners like F5, A10, Procera or ConteXtream are already present in the network, and Skyfire works closely with all of these and more to deliver a 60% boost in bandwidth. Other candidates for taking a discrete network application to the cloud include PCRF, OSS/BSS, content filtering, VOLTE, or orchestration of new service packages for a niche set of users. 

A second stage of NFV might involve an operator spinning up a "virtual full network" for a particular MVNO, or a nationwide enterprise customer that wants a private mobile network with its own policies, configuration and reporting. Only once that stage is proven and any scaling wrinkles are ironed out, can carriers then proceed to bringing their entire network in one region or nationwide to a virtualized architecture. So cloud-based QoE and optimization can be an important first step in that pathway-- and it can deploy today. It's already commercially proven.

Jeff Glueck

CEO Skyfire

Sign In