Networks, Automation, Connectivity & Data Centres · Chapter 9 of 18
The WAN: leased lines, MPLS, SD-WAN and what actually replaced what
How offices, factories and data centres connect to each other, and why a technology that dominated for twenty years is now a line item people are trying to remove.
Wide-area networking is the least glamorous part of infrastructure and one of the largest lines on an enterprise network budget. It is also where a great deal of infrastructure hiring happens, because every bank, manufacturer, retailer and hospital has branches, and those branches have to reach something.
The generations, and what each solved
| Generation | What it is | Why it appeared | Why it faded |
|---|---|---|---|
| Leased line (E1/T1, then Ethernet) | A dedicated circuit between two fixed points | Guaranteed bandwidth, guaranteed latency, no sharing | Expensive per Mbps, and point-to-point does not scale to n sites |
| MPLS | A carrier-managed private network with label switching and traffic classes | Any-to-any connectivity, real QoS, a single SLA and one throat to choke | Very expensive per Mbps, weeks to provision a site, and it backhauls cloud traffic the wrong way |
| Internet + IPsec | VPN tunnels over commodity broadband | A tenth of the cost, available anywhere | No QoS, variable latency, and managing hundreds of tunnels by hand is unpleasant |
| SD-WAN | Software control over multiple underlays at once | Use cheap links intelligently, fail over per-application, provision in hours | Still needs an underlay; the controller becomes a new single point of failure |
| SASE | SD-WAN plus security inspection delivered from the cloud | One policy for office, home and mobile | Vendor lock-in, and inspection points add latency if badly placed |
What SD-WAN actually does that MPLS could not
The core idea is simple and worth being able to state cleanly: separate the decision about which path to use from the physical link, and make it per-application and per-second. A branch has a broadband line, a 5G modem, and perhaps a small MPLS circuit. The SD-WAN device measures loss, latency and jitter on each continuously and steers each application to whichever currently meets its policy — voice on the lowest-jitter path, backups on the cheapest, everything else on the fastest.
It also does local internet breakout, which is the change with the largest practical effect. Under MPLS, a branch's traffic to a cloud application typically went across the private network to a central data centre, out through a security stack, and then to the internet — adding tens of milliseconds and a great deal of backhaul. Breaking out locally removes that entirely, and it is why SD-WAN and SaaS adoption are causally linked rather than merely contemporaneous.
The honest cost comparison, and the honest caveat
MPLS is commonly quoted at an order of magnitude more per Mbps than business broadband. That comparison is real and it is also not like-for-like: MPLS bandwidth is committed and comes with a latency and packet-loss SLA, and broadband bandwidth is contended and comes with an availability SLA at best.
The mature position — and the one that reads as experience in an interview — is that most enterprises now run hybrid: broadband or 5G as primary at most sites, a small MPLS or dedicated circuit retained at the sites where deterministic latency genuinely matters (trading floors, factory control systems, some clinical systems), and SD-WAN making the per-application decision between them.
“Rip out MPLS everywhere” is a consultant's slide. “Rip it out where the traffic does not need it” is a plan.
Last-mile options, and what they actually deliver
| Option | Typical delivery | Latency character | Where it fits |
|---|---|---|---|
| Fibre leased line | Dedicated fibre, symmetric, 1:1 contention | Low and stable | Primary at any site that matters |
| Business broadband (GPON/FTTH) | Shared fibre, asymmetric | Low, occasionally jittery | Primary at small sites, secondary at large ones |
| 5G / LTE fixed wireless | Cellular | Low but variable; can be excellent | Fast to deploy, good diverse backup, now a genuine primary at small sites |
| Cable / DOCSIS | Shared coax | Asymmetric upstream is the constraint | Common in the US, rare in India |
| LEO satellite | Starlink and similar | 20–60 ms — a step change from geostationary | Remote sites, disaster recovery, ships and mines |
| Geostationary satellite | VSAT | ~600 ms round trip | Legacy; unusable for interactive applications |
The design principle that matters more than the technology choice: diversity is not the same as redundancy. Two fibre circuits from two different providers that run through the same duct under the same road are one circuit with two invoices. Ask for the physical route, not the logical one — and pairing fibre with 5G gives you genuine media diversity for very little money, which is why it has become the standard branch design.
Check yourself