SD-WAN POC for Retail Network – Mind The Gap

When it comes to POCs, one of the more interesting presentations at the WAN Summit was given by Snehal Patel, network architect, at the Gap. The Gap, if you don’t know, has been sort of the poster child for Viptela’s SD-WAN. And while the needs of the Gap may not necessarily align with every small- to medium-sized enterprise, Patel does a good job providing a sort of “man on the ground” look at what’s happening with SD-WANs and how their POC worked.

Conducting a Proof of Concept (POC)

I won’t walk through the Gap case study in detail, but there were a few concrete takeaways that I thought were helpful. At a very practical level, if you’re looking to conduct a POC consider constructing a scorecard that prevents SD-WAN vendors from scoring well by dominating any one use case. Identify the use cases that are important to your organization and group them by area. Maybe ease of deployment and configuration are particularly important  for your end nodes. You’re use cases might be zero-touch deployment, policy configuration and the like.  Once you have your use cases grouped appropriately, weight them in their relative importance. If Zero Touch Deployment most important, give it a 5. If  policy is less important give it a 1. Only then grade vendors across each use case with the final score being the weighted result. As for purchase criteria, The Gap was looking to eliminate their underlying investment in routing. They wanted a policy-driven SD-WAN with a full routing stack (BGP and OSPF) and that could scale. Encryption should be IKE-less and with every location connected to multiple Internet paths, active-active with performance visibility was very important. Like many large organization, The Gap has learned about the importance of segmenting their backbones. The company wanted network segmentation with an L3VPN. Stores are widely distributed with little onsite expertise so Zero Touch Provisioning (ZTP) was important. With operations being done remotely, you’re looking at the need for easy code upgrade, circuit statistic and usage, centralized templates for configuration and centralized policies.

Broadband Not Always the Best

There’s the marketing of SD-WAN and then there’s the reality. The marketing of SD-WAN will tell you to connect your most important SD-WAN sites to two, dual-homed Internet connections. It’s complicated, but by connecting to multiple, diversely routed networks you can basically achieve the same theoretical uptime as an MPLS connection. (We walk through the math here if you’re interested.) The problem is that getting Internet connections, let alone diversely routed Internet connection isn’t always possible. Some locations will only be serviced by one supplier. If there are multiple suppliers their infrastructure will share common ducting or infrastructure. Stores in malls often face this challenge because often the contract for malls are awarded to one supplier. Patel talked these same issues in Gap’s network. Guaranteeing uptime for stores was a major challenge. Stores with common DSLAM or non-redundant providers were a problem as the site can lose both circuits at the same time, such as from a power outage. There were also connectivity problems stemming from how providers had configured their NATs.

LTE Secondary Connection

The Gap  addressed the problem by using LTE as a second primary connection. It’s the right play, if LTE is available in your region, which isn’t always the case. You always need to be ready for performance differences. Latency on LTE connections is significantly higher than other technologies, but jitter performance seemed inline with the other connections used in the deployment. Proper use of LTE may also need some basic RF engineering, depending on location. Common wisdom has it that MPLS is the better choice for a connection, but that’s not always the case. So much of it depends on the region, a point we noted in this post where I showed one of my customers how the Internet actually out performed MPLS going into northern Mexico.

Zero Touch Provisioning – Not with PPP Links

Patel touched on similar issues as well. In Europe, he noted that MPLS circuits tend to have a fair amount of packet loss. Some carriers also use the point-to-point protocol (PPP), which doesn’t work well with zero touch provisioning (ZTP). In the Asia Pacific, there are a number of Internet routing challenges in part related to The Great Firewall of China. I’ve seen similar issues as well, particularly in the Asia Pac. Am I being vague about the Asia-Pacific? You bet and so was Patel. And like him I’ll say, reach out to me directly if you’re looking for more information about that region of the world.  Check out our Ultimate SD-WAN Guide.