Bridging the Reliability Gap
MPLS carriers offer contractual SLAs because they control the entire path — their routers, their fiber, their peering points. The public internet is best-effort: packets compete for bandwidth, routers drop packets under load, and no single entity is responsible for end-to-end quality. So how does SD-WAN make the internet reliable enough for a CFO's board call?
The answer is a set of techniques applied at the SD-WAN overlay layer — a virtual tunnel (typically IPsec or DTLS) that runs on top of the physical internet connections.
1. Continuous Path Quality Monitoring
SD-WAN appliances exchange BFD (Bidirectional Forwarding Detection) probes or proprietary heartbeats every 100–500ms across every active tunnel. This gives the controller a real-time quality score for each path. When a path degrades, traffic is moved before users notice — not after they complain.
2. Forward Error Correction (FEC)
FEC is borrowed from telecommunications and storage systems. When the SD-WAN controller detects packet loss on a link (say, 3%), it instructs the sending appliance to add parity packets to the stream — mathematically derived redundancy data, similar to RAID parity on a disk array.
If the receiver gets packets 1, 2, and 4 but packet 3 is lost, it uses the parity data to mathematically reconstruct packet 3 without requesting a retransmission. This eliminates the round-trip delay of TCP retransmission, which is catastrophic for real-time voice and video.
FEC adds overhead (roughly 25–50% more bandwidth consumption when active), so it is only applied selectively to latency-sensitive traffic on degraded links.
3. Packet Duplication
For the most critical traffic — a CEO's video call, a trading system feed — SD-WAN can send identical copies of every packet simultaneously over two different physical links (e.g., Fiber and 5G). The receiving appliance accepts whichever copy arrives first and silently discards the duplicate.
This provides zero-loss delivery even if one link fails completely mid-call. The cost is doubled bandwidth consumption on the duplicated traffic, so it is reserved for the highest-priority application classes.
4. Jitter Buffering and Reordering
When packets traverse different physical paths, they can arrive out of order. SD-WAN appliances maintain a small jitter buffer — a short queue that holds packets briefly to allow reordering before delivering them to the application. This prevents the audio glitches caused by out-of-sequence voice packets.
Key Point: The reliability gap between MPLS and internet is real, but SD-WAN closes it through layered techniques: proactive monitoring catches degradation before users feel it, FEC reconstructs lost packets without retransmission delay, and packet duplication provides hardware-level redundancy for critical flows. The result is that most enterprises can replace 80–90% of their MPLS spend with broadband while maintaining or improving application performance.