• 12/02/2013
    10:06 AM
  • Rating: 
    0 votes
    Vote up!
    Vote down!

SDN Drives The Next Wave Of Innovation

SDN will profoundly affect the network communications market, creating chances for a few startups and altering revenue streams with a shift from hardware toward software and services.

Software-defined networking (SDN), unquestionably one of the IT hottest topics du jour, is a new architectural model designed to maximize networking efficiency, agility, and control. There are three fundamental characteristics within the SDN approach:

  • Separation of the control plane from the data plane
  • A centralized controller and view of the network
  • Programmability of the network by external applications

Since SDN detaches the control plane from switching and forwarding elements, it will make networks more programmable by abstracting network elements and exposing them to various applications. What could some of these applications be like? One use case of interest to a telecom carrier could be the ability to move application workloads efficiently from an enterprise to an operator's datacenter. For an enterprise, SDN can facilitate the hybrid cloud paradigm, allowing businesses to use public cloud services to manage peak enterprise workloads.

In addition, with the advent of big data, network data mining that collects information about real-time behavior of applications running on the network can be an SDN application of particular interest.

The adoption of SDN may be hampered by vendor inertia -- in other words, the strong installed-base skillsets from traditional network equipment vendors, which are only now ramping up their SDN portfolios. It will take time before IT staffs are trained and well versed in the new technology. However, increased pressure from the developer movement and agile application development should act as a counterweight to this market restraint.

For carriers, one big driver lies in capital expenditure and operating expense savings. Capex savings arise via the use of increasingly commoditized hardware. The control plane separation of SDN can be instrumental in delivering original equipment manufacturer switches built on merchant silicon, instead of purpose-built, proprietary appliances.

Moreover, network function virtualization efforts, geared toward decomposing networking functionality into pieces digestible by x86 servers, are well under way. From a management and orchestration standpoint, SDN provides an automated, comprehensive, and flexible configuration, which, coupled with centralized control and management, results in lower carrier opex.

[Need to learn the basics of SDN? Check out our slideshow, "7 Essentials of Software-Defined Networking."]

Discussions with large enterprises and Tier 1 service providers show that the SDN market is still in its infancy. The technology represents a generational shift in networking that will span close to a decade. However, some operators are already setting up proof-of-concept point-solution deployments. Large enterprises in segments such as finance and healthcare are also experimenting with SDN. Google is the most frequently cited example of an early adopter; it implemented SDN for its WAN via its new B4 network. The early assessment is that the point of inflection for SDN will occur in the second half of 2014 or, more realistically, in 2015.

SDN will be disruptive to the highly profitable networking industry. Networking currently is not very competitive -- compared to other IT areas, such as servers, storage, or even wireless infrastructure -- due to its vertically integrated, oligopolistic structure. The high switching costs create barriers to entry, while the distributed nature of networks spurs the need for superior interoperability, stability, and custom application-specific integrated circuits.

The flatter architecture of SDNs can improve interoperability while fragmenting the market by increasing choice at the hardware, controller, application, and orchestration levels. This will enhance competition and create opportunities for new players, and it will result in lower margins (comparable to those in the server or storage markets) for manufacturers.

Traditional network equipment vendors will have the opportunity to make up the potential margin loss due to hardware commoditization by monetizing the northbound application programming interfaces between SDN controllers and applications. They can also provide software such as network visualization tools or orchestration applications to help manage the entire network.

SDN will have a great deal of influence on the networking market and those who work with the technology. But with these changes, enterprises and carriers will gain more choice, flexibility, and network efficiency.


Cloud options?

What would be your advice for enterprises that are intrigued by SDN and want to try it in their datacenters, but without a big monetary commitment? Are there "SDN as a service" providers yet -- essentially specialists that abstract much of the complexity and let CIOs get a feel for what SDN technology can do for their companies?

Re: Cloud options?

It's my understanding that the service providers are behind enterprises in adoption of SDN, with the exception of Google. Because the users are actually leading the charge in SDN adoption, they have more expertise, in many cases, than the vendors or service providers. So probably they might be more likely to try it in a lab scenario. Ron, would you agree with that?

Re: Cloud options?

Susan, there already some service providers getting their feet wet with SDN. I would split the user groups in 3 categories: enterprise, carrier, and web 2.0/cloud providers (such as Google).  Google did their own thing (even coming up with homemade 10Gbit/s Ethernet switches). GOOG leveraged OpenFlow as the SDN protocol along with the Quagga software routing suite, which was already familiar to its team. But besides them, NTT Communications (talked a bit about them in my previous answer), Deutsche Telekom (TeraStream project for IPv6 in Croatia) and Colt Telecom (Ethernet Multi-Service Platform) are a few current implementations.  Then, there is Telefonica in Brazil (Virtualized customer-premises equipment - vCPE) which is initially mostly an NFV deployment but will eventually leverage SDN (I see NFV as a use case of SDN as the two technologies are complementary).  But for the most part, I would tend to agree with your obervation that the data center/enterprise opportunity represents the low hanging fruit for SDN and certainly it's one of the early use cases.  

Re: Cloud options?

@rgruia, thanks for your response. Great details on the specific carriers, which I have found difficult to find on my own.

Re: Cloud options?

Lorna, this is a very valid question... SDN as a Service is not quite there yet, but I'd say from the current service provider deployments out there, perhaps the closest to this is NTT Communications' Enterprise Cloud Service.  They have a global network of virtualized servers with virtualized network links within and among the data centers, with NEC's ProgrammableFlow controller being deployed in the data centers. This architecture enables the global network of datacenters to function like a single virtualized data center for scalable and seamless services.  So the end-customer reaps the rewards of the SDN architecture in this case, while NTT Comms gets CAPEX/OPEX savings, via the simplification of configuration changes, which become flexible and on-demand instead of requiring cumbersome setup and configuration procedures...  

Couple of thoughts about article and responses

I think the premise of the article is actually quite accurate, so kudos there. There is probably a little bit of precision required around some finer points. There continues to be a lumping together of SDN and commodity hardware. The premise is that separating out the intelligence allows for cheaper underlying hardware.

I don't think that point is actually necessarily true - at least not in the way people mean. Hardware prices will come down for sure - covergence on a smaller set of merchant silicon and competition will do that. But commodities happen when there is little or no differentiation. There is nothing inherent to SDN that limits differentiation. And the reality is that north of 80% of R&D costs are on the software side (and have been for more than a decade). The costs to recoup are not the small per-component hardware costs but the billions in R&D spend. Just droppin the price of components doesn't change that there is a ton of software effort going in. I believe the more likely outcome is a shift in pricing mix from hardware to software, with solution pricing being driven by competition more than any SDN technology.

On the question of enterprise vs service provider, the answer is more muddled mostly because people are still arguing over definitions. The basic framework of centralized policy management is not actually that new, so some people will claim they have been selling (vendors) or using (customers) SDN for umpteen years. 

In its current incarnation, though, the number of SDN deployments is small - in both enterprise and SP. R&E will be the first vertical to adopt (particularly OpenFlow). SPs have been aggressively pursuing SDN for awhile, but not only in OpenFlow form. Protocols like PCE (basically the algorithms that compute network paths) are potentially more interesting, as are protocols like ALTO (intelligent content routing based on proximity of content to user, for example). 

IMO, you'll see anyone who is working on cloud (public or private) move first, because the real problem being solved is workflow automation, and that is an integral part to providing on-demand services in an always-up cloud.

-Mike Bushong (@mbushong)


Re: Couple of thoughts about article and responses

@mbushong, thanks for your thoughts. Are you saying that cloud providers are already using SDN in their production networks? Or is no one really using it for business-critical uses yet?

Re: Couple of thoughts about article and responses

Some cloud providers are certainly using SDN (or SDN-like architectures). The extent to which you view them as SDN, derivatives of SDN, or predecessors of SDN depends a bit on whether you are a purist adhering to a definition, or a practitioner embracing principles. 

The most common SDN applications right now appear to be the Tap applications (steering traffic through monitoring points). Big Switch and Arista both feature Tap applications in their portfolio. HP has demonstrated (and has customer quotes around) Microsoft Lync in enterprise networks (I don't recall which enterprises though). I believe Indiana University and the University of Kentucky both have OpenFlow deployed in some measure (Indiana was referencing Brocade I think?). Smaller cloud providers have been adopting SDN as they build their business. Certainly Google's ambitions around OpenFlow are well-documented (but they are also looking at PCE very closely, which will be the more meaningful protocol for them IMO).

I mention all of this to suggest that there are lots of vendors already with kit deployed. The use cases tend towards more basic (not surprising). And those who don't have a ton of attachment to existing architectures are moving faster (so you will see smaller cloud service providers move quickly). 

Sorry for the long answer - I am missing my mental clarity this week. But one more thing - an interesting point here is that SDN largely solves a workflow automation problem. So the real question might be: are there cloud providers who are solving this problem well? EC2 might not meet SDN definitions in that their provisioning system is not controller-based, but they are absolutely going after the same problem space. I mention this because we could see a market where network virtualization, NFV, DevOps, and SDN all solve various elements of the major provisioning and management problems. It might not be just straight SDN or not SDN.

Re: Couple of thoughts about article and responses

@Mbushong, thanks for your response, and all the details. I agree that there are different ways to address similar problems, and I try not to get too hung up on definitions or terminology (so I sometimes oversimplify). I think the fascinating thing is to see the technology actually being implemented in real places like Indiana University and how it is changing the way people are able to work and function.

Re: Couple of thoughts about article and responses

Mike, you make a few excellent points here.  IMHO, the two things are feasible, i.e. a lowering of margins, say to something a bit more in line with other LOBs but at the same time, there will be a shifting happening from hardware to software and services.  It will be interesting to see how players like Cisco adjust to that (the early signs are that they are).  Bottom line is, if the future pie is bigger, even if the hardware% slice is smaller, if the overall pie is bigger, that could still be higher than today, but there will be two new components added to the mix.  Maybe before their pricing was built into the the total hw cost, while now it will become separate, but still... the modus operandi will shift to something of a hybrid between the status quo and an IBM Global Services type model.  So the traditional NEVs will have to understand the shift and embrace it.

But you saw the price/port for Insieme versus the older CSCO gear... the initial trend will be a drop in hw pricing, which is normal.  Of course, it goes w/o saying that they are very good at migrating a base away from the current to a newer technology (e.g. look at what happened to their switching/networking gear as CSCO got into IP PBXs last decade w/ its Selsius acquisition... initially, they were more focused on getting the data part of the deal and voice was only the gravy on top of that - w/ a pull-through effect anywhere from 3:1 to 5:1 depending on the existing network infrastructure).  One cannot under-estimate their ability to do that.