Software-defined networking is one of the two major pillars of network transformation that has emerged in this...
decade. When SDN was introduced, many expected it to sweep traditional routing -- and physical routers -- out of the network, but it hasn't. And while SDN has earned a place in the data center, SDN in the WAN is still rare. That could change, particularly as network operators begin to deploy 5G wireless networks. It's complicated.
Explicit control of forwarding via OpenFlow -- the Open Networking Foundation (ONF) version of SDN -- is the model most people think of when they think SDN. In the data center -- particularly in cloud data centers, where virtual LANs have to be deployed to build multi-tenant services and applications -- this version of SDN has been widely adopted. That use proves the model works, but it doesn't prove it can scale.
Scalability has been the biggest challenge with SDN in the WAN. If you ask whether the internet can be built on SDN, almost every network operator would be skeptical. If you accept that some IP networks are too big for ONF-modeled SDN to work, how big of a network would work? The answer is we don't know. And that's been the problem of SDN in the WAN.
The good news for SDN is 5G technology and other metro-focused changes in networking could build a bridge for SDN out of the data center. 5G has specific features that encourage SDN deployment, including network slicing). The combination of 5G, network functions virtualization, edge hosting, content delivery and streaming makes considerable new deployment likely in metro networks. New or greenfield deployment, particularly when in manageable metro-sized parcels, is the ideal place to introduce new technology, which could be the opening for SDN in the WAN.
Uses for SDN in 5G technology
Mobile networks have become a combination of mobility management and content delivery features. Emerging 5G networks add the need for network virtualization, and network slicing addresses the need for multi-tenant mobile and for hosting additional features to allow virtualization within those network slices. This opens two avenues to SDN in the 5G metro: using SDN for data center interconnect (DCI) and for mobility and content management.
Both 5G and mobile content are credible edge-hosting drivers, as service features and content cache points need to be close to those using the service. Edge hosting means network operators will use smaller data centers that are less efficient in handling variable loads, so we can expect them to create a virtual pool of edge-hosting points and deploy DCI on a larger scale. If fat pipes connect edge data centers that use SDN, it's logical to assume SDN in 5G technology would be extended across the DCI links to build resource pools across data center boundaries, which would extend SDN in the WAN.
If mobility management elements of the mobile network, like the 5G version of the Evolved Packet Core, rely on hosted features in SDN-connected data centers, it's fairly easy to see how SDN could be used to implement these elements. EPC technology today is based on tunnels that follow mobile users as they roam among cell sites. SDN forwarding could do the same thing, and the same SDN devices could provide direct connections between mobile content consumers and their cache points. SDN could support a rearchitected notion of mobility and content delivery that uses commodity white boxes, rather than custom proprietary appliances.
SDN could hit metro networks, or not
All of this means using SDN in 5G technology could explode in metro networking, and that mission is where most operator investment will be focused for at least the five years needed for 5G deployment. That happy news is tempered by two risk factors, and the result could easily turn out differently, leaving SDN locked in the data center.
The first risk factor that could relegate SDN to the data center is the pace and completeness of 5G adoption. Operators expect even radio-network-focused or 5G-fiber-to-the-node hybrids probably won't roll out in any volume until 2019 or 2020. Complete implementation of the 5G core is likely to take up to four years beyond that, and it's not yet clear just how far that implementation will go. The 5G standard for creating a hybrid 5G and 4G LTE wireless network is the first step toward 5G, but some think it may be the only step taken for a long time, unless major new service revenue can be defined to justify the broader 5G revolution.
The second risk factor is the emergence of network node operating systems like AT&T's DANOS and the ONF's Stratum, which could be used to support SDN, but could also host open, traditional routing and switching software. AT&T's decision to deploy 60,000 DANOS white box devices to host legacy routing software in cell sites shows these open white boxes might be the major risk to the role of SDN in the WAN and in 5G networks. White box legacy routers have the lower capital cost of SDN white boxes, but not the risks of deploying new technology.
ONF's Stratum may be a double-edged risk to broader SDN deployment in the WAN. The ONF launched the formal standards model of SDN. And now that ONF is championing an open-box OS option that could present an SDN alternative, it could be an indication that many early SDN supporters now doubt the technology will have universal appeal -- particularly the use of SDN in the WAN.
What might counter these risks? Operators say the biggest problem they have committing to SDN in WAN and metro buildouts is the lack of a mature and broadly accepted standard for SDN controller federation, which refers to east-west or between-controller APIs. Without this, large-scale SDN deployment requires either a highly scalable and available single-controller deployment or some as-yet-undefined higher-layer orchestration of SDN control domains. Initiatives have targeted controller federation for more than five years, but none has won broad adoption.
As network operators deploy 5G, IoT and other edge-hosting-intensive services, there's a window of investment for new infrastructure that SDN in 5G technology and the WAN could ride or be swept away. The long-neglected issue of controller federation may be the most important technical factor in deciding which of these possible outcomes will become reality.