Two networks, one on top of the other
An SD-WAN has two layers. The underlay is the set of real circuits and routers that move packets between sites: MPLS, broadband, DIA, cellular. The overlay is a second network built on top, made of encrypted tunnels between the SD-WAN edge routers. Applications run across the overlay. The underlay only has to carry the tunnel packets from one edge to another.
On small screens, swipe sideways or focus the diagram and use the arrow keys.
Overlays are not new. IPsec site-to-site VPNs and GRE tunnels are overlays too. What SD-WAN adds is that the overlay is built, routed and controlled automatically from a small set of central components, instead of being configured tunnel by tunnel.
On small screens, swipe sideways or focus the diagram and use the arrow keys.
On small screens, swipe sideways or focus the diagram and use the arrow keys.
Idea 1: transport independence
Because the tunnels run over any IP path, the overlay does not care what the circuit is. An edge with one MPLS line and one internet line forms tunnels across both, and the network sees two ways to reach the same site. The edge needs only an IP address on each transport and a route to the other edges' transport addresses.
On small screens, swipe sideways or focus the diagram and use the arrow keys.
- A new site can start with whatever circuit is ready first, then add another later, without redesigning anything.
- An internet circuit stops being only a backup. It can carry real site-to-site traffic, encrypted in the tunnel.
- Moving a site from one provider to another changes the underlay and leaves the overlay design in place.
Independence has limits. The overlay cannot make a poor circuit good, and it cannot bypass a provider that blocks the tunnel ports. It gives the edge a choice of paths and a way to tell the good ones from the bad ones.
Idea 2: centralised control and policy
In a traditional WAN every router works out its own routes and carries its own policy. In SD-WAN the edges report what they have to central controllers. The controllers decide what each edge should hear and send it back, along with policy such as which traffic may reach which site, or which path an application should prefer.
On small screens, swipe sideways or focus the diagram and use the arrow keys.
- Each edge connects to the controllers and reports its routes and its transport details.
- The controllers apply the administrator's intent to what they have learned, and send each edge the routes and policy that edge should use.
- The edges build tunnels directly to each other and forward user traffic between themselves. The controllers are not in the data path.
The last point is easy to miss. The controllers hold the plan, but the user traffic flows from edge to edge. That keeps the controllers off the path of every packet and lets one small set of them serve many sites.
Idea 3: path choice by measured quality
Edges send small probe packets through each tunnel and measure loss, delay and jitter. A policy can then say that voice needs a path inside certain limits and send it over whichever tunnel meets them. Where a traditional router sees an interface that is up, an SD-WAN edge sees a tunnel that is performing well or badly. A later lesson in this course covers how that works.
On small screens, swipe sideways or focus the diagram and use the arrow keys.
What a branch looks like with SD-WAN
Take BR-01 in the course lab, with system IP 10.255.2.1. It has an MPLS circuit and a business internet line. The edge forms tunnels across both to HUB-01 and, where policy allows, directly to other branches. The Controller (formerly vSmart) tells it which prefixes sit behind which sites, and the Manager (formerly vManage) is where the administrator defines the configuration and policy for the whole fabric. Lab names and site IDs here are examples, not requirements.
What SD-WAN does not change
On small screens, swipe sideways or focus the diagram and use the arrow keys.
- It does not remove the underlay. You still buy circuits, and still need working IP connectivity on every transport.
- It does not remove routing. The edge still connects to the LAN with ordinary routing such as OSPF or BGP, and the overlay has its own routing protocol on top.
- It does not guarantee quality. It can steer around a bad path, but if every path is bad, the user still has a bad experience.
- It does not do security on its own. Encryption of the tunnels is built in; firewalling and inspection are separate features you enable and design.
Mistakes when first meeting SD-WAN
- Saying SD-WAN is the internet replacing MPLS. It is a way of using any transport together. Whether MPLS stays is a business decision.
- Thinking the controller carries user traffic. It carries control information only.
- Mixing up overlay and underlay addresses. The tunnel endpoints use underlay addresses; the user prefixes live in the overlay.
Test yourself on the SD-WAN idea
Which description best fits the overlay in an SD-WAN?
A branch gains a second line from a different provider. What does transport independence mean for the design?
In a working SD-WAN, where does user traffic between two branches normally flow?
An engineer says SD-WAN means the provider circuits no longer matter. What is the best correction?