SD-WANs: A New Beginning for Network Engineering

For decades network engineering was the cornerstone of, well, everything. Without a predictable and reliable network, businesses stumbled, and network engineering prevented that from happening. No wonder Gartner 2017 IT Key Metrics Data (access required) recommend enterprises allocate 13.4% of their total IT costs in 2017 towards networking with WAN support consuming the greatest percentage of their network support budget. 

Like so much of IT, though, network engineering grew out from our need to overcome constraints. With most cases we’ve been able to use technology to overcome our limitations. Moore's Law finally caught up with business computing requirements giving us "just enough" processing power. 

The same was true with storage. We held our breath until storage capacity reached the per square inch and I/O operations per second that could meet our requirements.

But no constraint has been as persistent networking. Getting high-capacity, high-quality and affordable connectivity at a global scale has long eluded businesses.

Network engineers made the best of this bad situation.

Network engineers squeezed every cent from our MPLS capacity, calculating optimized routes to global or regional data centers. Optimization appliances were deployed to squeeze even more out of those thin MPLS pipes. And we added inexpensive Internet links, not MPLS links (when possible), in case a failure should come when dealing with a flurry of configuration changes. 

And they did all of that using obscure command line interfaces (CLIs). It was bad enough that MPLS was expensive but precisely because of the lack of bandwidth, MPLS operations became challenging.

Elimination of that complexity became a driving force behind traditional SD-WANs.

The SD-WAN proposition is enticing: aggregate cheaper Internet capacity, easily deploy it, and get rid of legacy MPLS networks. SD-WANs build an overlay across services, masking their specifics from applications. By dynamically routing traffic based on link quality, availability, and application requirements, SD-WANs make multiple data services appear as one single network.

And with fat Internet pipes, SD-WANs avoid much of the engineering needed with MPLS.

Network engineering is further minimized because SD-WANs simplify and automate much of the traditional network operations. Rather than achieving application delivery by configuring devices and working through redundancy configurations, SD-WANs allow IT to specify tolerances and objectives and leave the details to the SD-WAN.

SD-WAN seems like the answer to IT’s dream.

Inexpensive bandwidth, predictable delivery, global scale, and it frees up valuable engineering resources. From a cost-benefit standpoint, customers gain from WAN agility and avoid MPLS costs.

What more could IT want?

But to replace MPLS, SD-WANs must extend globally -- without relying on the Internet. . Despite being around since the dawn of large scale computing, the Internet was not designed to be the ultimate global networking platform. It's loosely coupled design, a collection of public and private entities with often conflicting interests, has promised "best effort" networking - at best. For many businesses, “best effort” isn’t good  enough.

Within well-developed Internet regions, these limitations might be less noticeable. SD-WANs can avoid convoluted routing, high packet loss at carrier exchanges and more, by selecting another route or service provider. But send traffic across long distances, between Internet regions, or areas serviced by only one Internet provider and traditional SD-WANs stumble.  The predictability gap between the Internet and MPLS is just too great. SD-WANs end up augmenting, not replacing, legacy MPLS, making their immediate hard ROI far less obvious.

SD-WANs need to extend globally.

To take on MPLS services, SD-WANs need to extend globally, providing routing control, packet loss and latency guarantees everywhere.

This is not an easy task to accomplish because it requires a new type of carrier -- one that provides affordable, predictable, global connectivity. But done right (and married to an SD-WAN), enterprises gain a unified network with tremendous amounts of bandwidth and optimal access to all resources: WAN, Internet and Cloud.

We can use any ISP, give branches however much capacity they require, connect to any resource - and eliminate MPLS. So many challenges addressed by network engineering can get solved, automatically.

Network engineering is moving out of the black hole of grunt work like many IT disciplines that migrated to the cloud. While the days of CLI might be coming to an end, the modern enterprise faces enormous challenges in the areas of:

• Optimal access to cloud applications
• Securing and integrating cloud infrastructure
• Enabling business continuity for a highly distributed business

 Surely, this will be a large enough task for network engineers that will yield significant return for the business.

About the Author

Dave Greenfield is the secure networking evangelist at Cato Networks. He brings more than 20 years of experience in IT and telecoms having worked as an award-winning journalist, blogger, and a technology analyst advising companies on their IT and WAN strategies. Dave is the author of the “Ultimate WAN RFP” and the “Essential Guide to Optical Networks.” He has a background in philosophy and computer science.