SDN architectures

5 Opportunities for NFV in the 4G Core Network

The virtual EPC (evolved packet core) was one of the standout network technology themes from this year's Mobile World Congress, with Intel Corp. (Nasdaq: INTC), IBM Corp. (NYSE: IBM), NEC Corp. (Tokyo: 6701), Ericsson AB (Nasdaq: ERIC), Cisco Systems Inc. (Nasdaq: CSCO), and Nokia Networks all showing demos of various kinds. (See: Packet Core Looks 'Ripe' for Virtualization and V Is for Virtualization.)

Paraphrased from my interactions with a number of mobile network operators, I’ve outlined why the virtual 4G core is viewed as a viable and attractive target for network functions virtualization (NFV) vendors, and why it makes sense for mobile network operators.

  1. Reduced opex: By replacing an array of proprietary and semi-proprietary platforms, and standardizing on a common hardware platform (x86 commercial-off-the-shelf servers), operators aim to reduce overheads and streamline support and maintenance teams. In the same vein, by standardizing on high volume servers, operators can potentially accelerate time-to-market and not have to buy brand new, proprietary equipment that is three years out of date the day it's installed.

  2. 4G is an expanding deployment: Operators are interested in NFV for new and expanding deployments, but less enthusiastic about re-writing legacy applications. But LTE and the Evolved Packet Core are new deployments in growth mode in virtually all parts of the world. And, just as importantly, the services and functions that orbit the 4G core (optimization, analytics, DPI, security, etc.) are very well suited to virtualization and NFV.

  3. Mobile user-plane is "affordable:" Data traffic and end-user speeds in LTE are increasing, but data-plane throughput in mobile networks remains trivial compared to fixed line broadband. This makes the 4G data-plane relatively “affordable” in processing terms (as one operator executive put it), and makes it possible, and even attractive, in some cases, to run Serving Gateway (S-GW) and Packet Data Network Gateway (P-GW) applications on server platforms. Moreover, some mobile core network elements have always been transaction-orientated, and, as a consequence, run on server platforms based around Intel processors.

  4. Service agility: A software-centric network environment is inherently more programmable than a “classic” hardware-based core network. The hope is that this ability to launch and modify functions more quickly can be translated into "service agility." New language is emerging around this idea, with terms -- such as “service-chaining," "VNF itineraries," and "virtualized service matrix" -- emerging to communicate how a workflow and processing path can be created in software according to a user's service definition.

  5. Open for independent software vendors: Moving the core network and related service platforms to a cloud environment will help open up the carrier environment to more dynamic software developers. "We could easily consume all the support resources of a small 100-person vendor," said one operator executive I spoke to recently. "Software makes it possible for smaller vendors to write to our environment," he said. "They just click into our datacenter."

There are a lot of reasons to recommend the development and implementation of a virtualized 4G core, and it feels inevitable that virtualized EPCs will be deployed at some point in the future. However, there are also a number of significant barriers, which I'll address in my next column.

— Gabriel Brown, Senior Analyst, Heavy Reading

Page 1 / 2   >   >>
Dredgie 9/30/2013 | 2:42:42 PM
Re: distributed and virtualized It's a good topic in general. As you say, there's a ways to go, for sure. Supporting the ISG (as they are not a standards body) there will be a second BOF on the topic (NSC as was – now Service Function Chaining / SFC) at IETF 88 in Vancouver. But according to the area Directors, if the meeting had been a few weeks later, they would have gone straight to working group status, as is the large number of contributions that have been developed, on the topic, under the auspices of this WG. They do have to agree on a charter as well, of course! :-)
Gabriel Brown 9/30/2013 | 1:46:36 PM
Re: distributed and virtualized Thanks Dredgie. The idea that EPC (and related) functions are decomposed and then reassembled into a service chain is a powerful one. There are far-reaching implications for how application software is written -- and it may be harder to square this with a "quick and dirty" port of an application to COTS. In short, vendors are going to have make quite a substantial investment in re-designing software for a virtualized deployment.

I should probably have made this one of the 5 Challenges for NFV in the 4G Core Network.
Gabriel Brown 9/30/2013 | 1:35:37 PM
Re: Innovation acceleration - the sleeper? Thanks Charlie. Rapid innovation certainly appears to be at the heart of AT&T's Domain 2.0 strategy announced recently, which seeks to decouple network hardware and software. Obviously there are many aspects to rapid innovation, as you mention in your post.
charlieashton 9/30/2013 | 11:26:49 AM
Innovation acceleration - the sleeper? It is worth emphasizing the "innovation speed-up" benefit to the service agility concept that you mention. As well as accelerating the deployment of services, SDN/NFV also reduces the risk and costs associated with the development, testing and field trials of totally new services in the market.  Service providers can simply instantiate them as virtualized applications rather than having to procure and deploy new, fixed-function hardware which may not be easily reusable for other applications. Also, using the service chaining concept you mentioned (still in its infancy), makes it possible to create "service mashups" that are combinations of previously separate applications (including more applications developed by third parties and partners) – another form of rapid innovation.  To many, rapid innovation is the key to carriers increasing ARPU, competing effectively with the OTT players and being more than just 'data pipes.'
Dredgie 9/12/2013 | 11:04:50 AM
Re: distributed and virtualized I guess it's decomposed, distributed and virtualized if you consider the EPC elements: policy, state management, L3 forwarding (decomposed RIB / FIB with TBD protocol to inject routes – ForCES, maybe?) and L7 steering – specifically through the 7+ different service chains mobile operators would expect to have (comprising LBs, DPI, IDS, IPS, FWs,  Caches and the like.
lrmobile_janly0 9/11/2013 | 2:16:00 PM
Re: distributed and virtualized
Yes it may be more distributed (logical/physical would be a matter of 
implementation) to adapt to the underlying transport infrastructure capacity and topology. This would bring a performance advantage compared to a purely centralized model (few sites) on top of the advantages you describe above (e.g. reduced costs). I think this could help solve the signalling overload and data crunch problems.
Gabriel Brown 9/11/2013 | 4:13:14 AM
Re: Making waves (or is it wiggles?) in the EPC with CloudNFV Hi Dredgie. Thanks for the comment. I'm not familiar with the "wiggle", but...

There is the notion of a "virtualized service matrix" emerging, but that's more on the Gi side of the P-GW (although in time will incorporate the EPC profile as well)

In essence this is the ability to program "workflows" in software according to the use-case or customer type. The operator can program virtualized instances of the EPC (and related services) according to different service definitions, which incorporate traffic models, mobility profiles, policy and charging, security requirements and so on.

This configuration stays with the user (or user group) as she/he moves around ("wiggles"?)

In theory, it gives operators scope to diversify their service offers and address a greater part of the market more quickly. In practice, operators are not ready to offer such 'granular' services. They want to keep things simple to start with.
Gabriel Brown 9/11/2013 | 4:01:18 AM
Re: distributed and virtualized Thanks gianlu. Do you mean an EPC that is logically and physically distributed?

With virtualization there is also now a notion of physically centralized, but logically distributed EPC (i.e. running in the data center but subtended to a fewer number of sites).
lrmobile_janly0 9/10/2013 | 5:17:51 AM
distributed and virtualized
Good points. I would also point out that a 4G Core may be both
virtualized and "distributed" to route traffic more efficiently compared
to the existing centralized model.
Dredgie 9/9/2013 | 6:23:57 PM
Making waves (or is it wiggles?) in the EPC with CloudNFV On the topic of agility (flexibility) and efficiently, Tom Nolle talks-up a nice, hypothetical, EPC infrastructure service that could be enabled under CloudNFV. I'm sure I'm doing this a complete disservice, but with the ability to define anything as an abstraction – including service connections (i.e. tunnels and lines) - you could create something he refers to as a 'whip': fixed at one end while moving at the other. In the EPC case, you can create a tunnel that is fixed to the PGW but waves (or wiggles!) through cell sites at the other end. This capability is provided at the service level without being tied to a specific implementation, as you would typically tie mobility management to a PGW/SGW pair today.
Page 1 / 2   >   >>
Sign In