The Old Way: Backhauling
Imagine every employee at a company's 50 branch offices needs to get their mail delivered. But instead of each office having its own mailbox, every piece of mail — even a local postcard — must first be driven to headquarters in another city, stamped by the central mail room, and then driven back. That's essentially what traditional enterprise networking did with internet traffic.
For decades, companies connected branch offices using MPLS (Multiprotocol Label Switching) — a private, carrier-managed network technology that creates dedicated, labeled paths for traffic between sites. MPLS operates at Layer 2.5 of the OSI model (between Layer 2 switching and Layer 3 routing), using short fixed-length labels instead of IP lookups to forward packets at high speed along pre-determined paths.
Pros of MPLS:
- Guaranteed performance with a contractual SLA (Service Level Agreement) — if the provider drops packets, they pay penalties.
- High privacy: traffic never touches the public internet.
- Predictable, low-jitter paths ideal for voice and video.
Cons of MPLS:
- Extremely expensive — often $300–$500/month per megabit, versus $1–5/month per megabit for broadband.
- Slow to provision — new circuits take weeks or months to install.
- Rigid and inflexible — scaling up requires renegotiating contracts and waiting for physical installation.
The "Trombone" Effect:
The deeper problem wasn't just cost — it was topology. MPLS networks were designed around a hub-and-spoke model where all traffic flows through a central headquarters. In a world where most applications lived in the data center, this made sense. In a world where employees use Salesforce, Microsoft 365, and Zoom — all cloud-hosted — it creates a catastrophic inefficiency.
If a user in the London branch wants to open a Salesforce record, the traffic path looks like this:
- London Branch → MPLS Circuit → New York HQ Firewall
- New York HQ → Internet → Salesforce (San Francisco)
- Salesforce → Internet → New York HQ → MPLS → London
The London user is experiencing the round-trip latency of New York plus the latency to San Francisco, all to access a cloud app that could have been reached directly from London in milliseconds. This is called backhauling — dragging traffic backward through the network before it can go forward.
As enterprises moved to SaaS and cloud-hosted applications, backhauling became untenable. The MPLS circuit — already expensive — was now being saturated with YouTube streams and Teams calls that had no business traversing a private WAN. Something had to change.
Key Point: MPLS was designed for a data-center-centric world. When applications moved to the cloud, the hub-and-spoke model turned every branch office into a latency and cost problem. The trombone effect is the root cause that SD-WAN was built to solve.