Taming the Wild, Wide WAN: Applying SDN to the Wide Area

Stephen Vogelsang is VP of Strategy, CTO, Nokia IP/Optical Networks (ION)

Networks should be responsive to the needs of users and applications. And yet, for a very long time, they have not been. They have forced the users and applications to adjust to network limitations. In the enterprise, IT departments have struggled to align the security capabilities of the network with the evolution of devices and applications. And in the wide area network (WAN), bandwidth cost and limitations have forced users to accept lower performance than the local area network (LAN).

All of this is changing because of the cloud. The promise of the cloud is to provide users with the very same immediacy and responsiveness that they would experience on the personal computer in their work office, from anywhere, on any device — over the network. In order to live up to this promise, cloud applications running in data centers have required the data center network to adapt and meet the cloud’s tremendous need for flexibility, scalability and agility.

As the cloud computing model expands, there is increasing pressure for the WAN to do the same. Thus, it is no coincidence that the principles for architecting data center networks have turned out to be the same principles that now need to be applied to the WAN.

Software-defined networking (SDN) was born in the data center as the only possible response to a cloud that ran on software-defined virtual servers. The cloud promise — to meet the needs of its users at any time — called for an architecture that could adapt to massive shifts in user demands. To match these shifts, the DC network had to be equally flexible, agile and scalable.

It also had to be automated. As users demand resources, the cloud spins up virtual machines (VMs) in the blink of an eye. The network must respond just as quickly to match it, configuring its own resources to handle the routing required in order to meet the connectivity needs. To make sure this all works without human intervention, which would be far too slow, the cloud and the network have to be able to talk the same language.

Where IT might see users, apps, domains and zones, the network team would see virtual local area networks (VLANs), subnets, switches and routers. Where IT would see issues of policy, rights and compliance, the network team would see access control, firewall rules, classes of service and SLAs. The challenge for SDN was to use software to automate and abstract the network using the native language of IT.

The results were dramatic, leading to the development of an entirely new networking paradigm. The IT view of the world is instrumented in policies that the SDN controller translates and binds directly to the workloads in software on the server, which instantiate the needed network connectivity and policies. Compliance issues disappear and are easily verified because they are built into the policies. And, the network fabric or “underlay” is dramatically simplified because connectivity policies are implemented in software. This makes it easy to grow the network to stay ahead of the scale out performance demand of the cloud. The abstraction provided by the SDN layer decouples the network fabric from application policy and allows them to scale independently.

With the decision to build cloud principles into the 5G networking standards, as well as the rise of the internet of things (IoT) and the spread of automation in general, the cloud is moving beyond the data center, being distributed to the WAN edge. Network functions that previously ran on dedicated hardware and applications and might previously have stayed in the data center are being located in edge clouds. Automation of machine processes, for instance, requires very low latencies. This results in workloads in the edge clouds that are critical components of the WAN or have stringent performance requirements from the WAN.

Somehow we need to automate connectivity and policy for these new workloads both within the data center and across the WAN. Yet the WAN is a very different beast. We cannot simply extend the data center SDN to the WAN because the edge cloud workloads require a binding to the underlying transport network.

The WAN is a collection of established, complex and global network infrastructures, allowing someone on a remote island in the Indonesian archipelago to dial an ancient rotary phone and connect with someone in New York City talking on the latest smartphone over LTE. There are layers upon layers of fiber optics, gateways, routers and switches, as well as operating systems and business systems that make this magic happen with 99.999 percent reliability. Empowering flexible IT applications of the cloud age to interact responsively and constructively with the wild, wide WAN is a much bigger deal.

Fortunately, WAN operators have already begun embracing some of the principles of SDN and virtualization, simply because it is a more efficient way to build networks. But it is a long way from these initial steps to anything that resembles today’s fully automated, software-defined, policy-driven networks.  Similar to the data center SDN, the WAN SDN software uses a language to abstract complex network devices and topologies, automate device configuration and optimize network resources.  However, the language is different and the WAN SDN does not have visibility into the workloads running in the data center.

There have been a number of attempts to use routers or switches that speak both data center SDN and WAN SDN languages as translators or gateways between the WAN and data center LAN, but they have taken a lot of integration effort and ended up being vendor-specific solutions. They have also tended to provide limited visibility of the end-to-end network services and have relied on protocols, such as RSVP, which are not designed for fine-grained control of network resources.

What the industry needs is an open and scalable way to abstract the WAN, which is multi-vendor, provides fine-grained control of end-to-end network services and ties seamlessly to the data center without complicated translators or gateways.

There are a number of new innovative technologies filling in pieces of the puzzle such as segment routing and open object models, and no lack of continued investment.  We’ve automated the data center LAN using the native language of application workloads and now the WAN is more programmable.  Watch in the coming months as the missing puzzle pieces are created and we complete the picture required to realize the full promise and power of the networked cloud.

Opinions expressed in the article above do not necessarily reflect the opinions of Data Center Knowledge and Informa.

Industry Perspectives is a content channel at Data Center Knowledge highlighting thought leadership in the data center arena. See our guidelines and submission process for information on participating.