SDN architectures

SDN: A Leveler for CSPs

According to data growth statistics, over the coming years we will experience a tremendous jump in data generation and consumption. Cisco forecasts that global cloud traffic is expected to grow 4.5-fold. That amounts to a 35% CAGR -- from 1.2 zettabytes of annual traffic in 2012, to 5.3 zettabytes by 2017. Overall, Cisco expects global data center traffic will grow threefold and reach a total of 7.7 zettabytes annually by 2017.

If those numbers seem impressive, consider that IDC estimates that the Internet of Everything will amount to an $8.9 trillion market in 2020. To co-ordinate IDC's estimated 212-billion things, data centers will need to shoulder the burden. This clearly puts the onus on networks to support the enormous volume of data that is being transmitted across channels. As customers' demand for bandwidth increases, there is a need for a strong network infrastructure at the back-end and intelligent devices to minimize the service latency and support this data growth.

Within this context, SDN is likely to reshape the telecom industry in new and interesting ways. Telco CTOs/CIOs worldwide are facing enormous pressure on IT spending, which will have an impact on their SDN strategy. However, they cannot ignore the tremendous benefits SDN would bring in terms of agility, flexibility and cost savings (both capex and opex), as enterprises are shifting from physical infrastructure to cloud based infrastructure.

Reducing costs and improving network efficiency
With SDN, carriers can increase the flexibility and manageability of their networks, helping to deliver bandwidth where and when customers need it with a (more) self-serve model. Operators can potentially improve both their network flexibility to improve customer value, while also reducing the high operational costs of supporting OTT providers such as Google, Amazon and Skype, which otherwise could challenge carriers' ability to grow their revenues and affect their margins.

SDN would provide the bandwidth flexibility that will give carriers the edge they need to successfully deliver a range of cloud-based services. In particular, SDN will help carriers to:

  • Automate the provisioning and maintenance of network services, which are highly people-intensive in the current scenario. For example, some estimates suggest that enterprises could see up to a 95% reduction in capital costs by implementing commodity silicon within switching infrastructure.

  • Streamline multi-tier architectures into a flatter, fabric-based infrastructure, so reducing operational costs associated with real estate, power, cooling, cabling and a host of other factors.

  • Automate traffic management, improve bandwidth engineering and enable CSPs to tailor the network "on demand" to customer needs. Today, large carriers are unable to offer customers truly flexible data networks, where customers can pay for only the bandwidth they use and automatically burst traffic as applications require.

    Building bolder business models with SDN and NFV
    SDN with NFV offers a way for CSPs to break out of their established, inflexible business models. CSPs can develop flexible models that are based on tiered services and on elastic services (e.g. security-as-a-service, unified communications-as-a-service, etc.) They can also pay network equipment providers on a per-use basis for network functions as needed, as opposed to capital expenditure tied to under-utilized equipment.

    The SDN-driven environment will not only help incumbent providers become more agile and able to adapt to market trends and subscriber demands more effectively, but will open up the market to Tier 2 and Tier 3 players that may not have had the deep pockets needed to develop proprietary hardware. It will allow new carriers to quickly scale and compete, as they won't have to load up on costly central office equipment to get started.

    Tier 2 and Tier 3 CSPs will benefit from:

  • Shorter time to market and lower costs for new services through running production and test and reference facilities on the same infrastructure, and not using commoditized hardware
  • More choice and less risk as pure software competitors emerge, due to the fact they don't need to invest in hardware anymore
  • Near real-time resource allocation to ensure network resiliency and protection from failures
  • Multi-tenancy support allowing carriers to securely share hardware resources
  • Improved power management; servers can be switched off or go into low power mode during non-peak hours
  • A competitive advantage against OTT services through easier integration of third-party services

    The early bird gets the worm
    SDN adoption is still in its infancy. While some carriers have realised early adoption can substantially benefit their organization, others are taking a cautious approach due to the capex required and the negative impact on the company's competitiveness associated with the immature product. Yet as the technology matures, we see most operators turning to SDN to protect their future and benefit from lower cost of operations, new revenue streams and strengthened network services. Failure to plan for SDN now, could put CSPs at risk tomorrow.

    — Ravi Kumar Palepu, Head of Telco Solutions, Virtusa Corp.

  • VictorRBlake 11/5/2014 | 3:01:43 PM
    Agreed Mitch,

    I absolutely agree that opex is a huge factor. I thought that was my principal point, that SDN doesn't show any promise of reducing opex in the short or mid-term because it adds complexity and requires a huge (operational) cost of deploying and integrating new technology, training, etc.

    All of that is with the promise of greater network flexibility in the future. The truth is that today's IP networks are very flexible. All of the applications we've dreamt up can and do work on it. As I said, there are a lot of reasons to want to virtualize networks for various marketing, security, or other virtualized needs. These were osentensibly the driving factors to create SDN.

    But I think everyone is still hard pressed to say "here is something we cannot do with IP today." Or as you argue perhaps too difficult to do with IP today. An example of that might be VRFs/VPLS which is largely manual today. So I get how SDN can potentially automate that. But as I said, there are huge chunks of work missing that would be required to do that because today the ONF specs give us just flow tables without distributed decision making required for security and many applications.

    Even when it is done, I expect that opex will not decline, it will rise with the complexity. Today the SDN platforms require way more work to deploy than off the shelf Ethernet/IP gear. You have to install an OS, patches, and then the SDN/ONF app, then tools, all just to configure a controller -- which still doesn't give you an operational network. You then need interop between the controller and the device, etc.

    I think we (all ?) agree that SDN holds great potential to accellerate application development for service providers who want to be in the mix. But the change isn't just one for sp's -- its about SPs and apps working together and that isn't going to reduce any costs at all. It is going to raise cost and complexity -- albeit with perhaps some fascinating benefits to application performance, etc. And those will likely be worth it. But going in thinking it is going to reduce $ seems to ignore both history and reality as it stands.
    Mitch Wagner 11/5/2014 | 2:48:31 PM
    Re: Flexibility sure -- cost - no evidence whatsoever Capex, for reduced upfront spending, is only part of the equation. SDN also promises to reduce opex, or the ongoing cost of ownership.
    Carriers are most interested in increased flexibility for increased revenue rather than reduced cost.
    VictorRBlake 11/5/2014 | 10:15:06 AM
    Flexibility sure -- cost - no evidence whatsoever While it makes sense to argue that SDN will over greater flexibility (aka the programmable network), there is no evidence to suggest that it lowers CSP costs at all. I know the academic argument is that commodity hardware will lower the hardware cost, but that is hardly the whole picture. Even if capital costs are dramatically lower there will be massive expenditures to build controller and OSS infrastructures as well as test and deploy all new equipment. In both the short and intermediate term these do not lower costs, but raise them.

    In DCO there is a statistical advantage to SDNs with near 100% port utilization that is definitely advantageous. Although high (port) utilization was possible with VLANs, SDN extends the concept beyond Layer 2 virtual networking to IP, MPLS, and now even PBB tags and labels. But port utilization will not fundamentally alter the economics of transit and access networks because port utilization is not based on (application and network) convergence and high densities. Port utilization in access and transport networks is instead based on geographic density which SDN offers no benefit nor (nor does it claim to).

    Far from being a "leveler" SDN is likely to increase the barrier to entry for CSPs as it increases the capital, engineering, and development costs in the immediate future.

    Where SDN levels the field is not for CSPs, but instead for applications. Whereas the construction of VPNs and other network operations have sometimes been required to support novel network applications, SDN offers (but by no means guarantees) the potential for rapid deployment of flow based alternative models of network operations. It's hard to say (outside the data center) what the "killer app" for SDN will be -- if there are sufficent improvements in and investment in the technology -- it is possible that broadcast could become a killer app, but many technologies have tried to conquer broadcast over packet switched networks and failed primarily because of inter-AS challenges.

    I just don't see the evidence that SDN levels anything for CSPs. I don't see little players benefiting from SDN. If anything it appears the opposite is true, the technology threatens to display big players with bigger players.

    One thing is certain. SDN is not "simpler" than "legacy" distributed IP and Ethernet technologies in any way. Because the technology must be backwards compatible it is 100% as complex as what is deployed now plus the addition of numerous layers of complexity. Adding layers of complexity doesn't decrease cost (aka lower the barrier to entry). It raises it. Does it increase flexibility -- yes it could -- it might -- if the OSS modifications can be made. But history tells us this is a long and costly process.
    Sign In