Rarely do I have a conversation about networking when the topic of software-defined WANs (SD-WANs) does not come up. It’s, far and away, the thing that network professionals care most about, even ahead of data center SDNs.
In a data center, enterprises can still people their way out of problems as there’s always an abundance of experienced engineers locally available to tackle any issue big or small. That luxury does not exist with the WAN as branch offices can be scattered across the globe and often, the best one can hope for in terms of a local resource is a branch administrator or someone who can check lights or confirm things are plugged in and powered up.
Also, for many geographically distributed organizations, the WAN IS their business – so having an agile, dynamic WAN that enables applications to perform better is a top priority.
However, the journey toward deploying an SD-WAN may not be obvious. The known end state is somewhat easy to envision – a fully autonomous WAN where applications performance is akin to a local LAN, security is integrated into the network and the network has the same level of agility as something comparable to containers. Sounds great, but how the heck does one get there? It’s like self-driving cars. The long-term vision is hopping into a car, where there’s no driver, steering wheel or any kind of conventional controls. The passenger simply says, “take me home” and then gets to work, watching a movie or taking a nap while the car goes off on its merry way automatically correcting for traffic congestion or other unforeseen issues. Seems simple, but where do we start?
My advice to network professionals has been to think of the journey as a series of chip shots instead of a single monumental moon shot that’s so intimidating that it seems unreachable. This requires thinking of an SD-WAN as a set of smaller tasks, instead of one big project and doing those tasks a step at a time instead of trying to achieve the utopian state overnight. Here are my top ten recommended steps on how to proceed.
- WAN optimization. There’s no sense in moving to an SD-WAN if the network has not been optimized. Without that, it’s tough to understand traffic patterns and bandwidth requirements.
- Visibility tools. You can’t secure or manage what you can’t see. Visibility tools will shine a light on a few existing blinds spots like what cloud apps are being used, how fast the cloud is growing, where multi-media traffic is going from and to, etc.
- Any combination of transport, including broadband. Most network engineers aren’t ready to say, “no mas” when it comes to MPLS. That’s fine, but move to an active/active hybrid WAN where the broadband and MPLS can be fully leveraged together.
- Dump the router. A super expensive branch router is no longer needed to have local routing capabilities. Replace that router with an SD-WAN appliance or anything else that can run a virtualized router, such as a WAN optimizer. Another option is to move routing the cloud.
- Local internet breakout. Having cloud traffic backhauled through a central hub is a horrible waste of bandwidth and ultimately impairs the application user’s experience. Instead, having the ability to identify known and trusted cloud applications in branch offices and connecting users directly to the Internet, over any combination of connections installed in step three above.
- Virtualize firewalls. Local Internet breakout requires localized security but that doesn’t mean having to deploy six figure firewalls at every location. Instead, run a virtualized firewall on the WAN edge device that already supports routing. Alternatively, internet-destined traffic can be directed to a cloud-based security service.
- SD the branch. Now that firewalls and routing have been virtualized, it’s time to virtualize other branch services such VPNs or other security or management tools.
- Dynamically mesh. This is getting pretty advanced, but over time, orchestration tools can be used so peer-to-peer connections, like the ones required for video, can be created dynamically. The tools exist to do this today, but it does take some advanced engineering to accomplish. It’s well worth it though.
- Analyze and optimize. Collect data, analyze it and tweak steps 1-9. Network traffic changes all the time and the network must be tuned periodically.
- Retire MPLS. This can actually be done much earlier in the process, but I’m placating the conservative folks reading this. Over time, multiple broadband connections, optimization tools and virtualized network services can replace MPLS and give comparable or better application performance at a significantly lower cost.
(This article first appeared here).
