SD-WAN path selection, SLAs and application performance
In short: SD-WAN recognises applications, constantly measures every link, and sends each application over a path that meets its service level. If a link degrades, traffic moves within seconds, and features like forward error correction and packet duplication keep voice and video clear.
Knowing the application
Path decisions start with knowing what the traffic is. SD-WAN edges identify applications by ports and addresses, by DNS and TLS information, and with deep packet inspection signatures for thousands of common apps. Traffic is then grouped into classes such as real-time voice, video, business-critical SaaS, general web and bulk transfers.
Measuring the paths
Each edge sends small probes through every tunnel (often using BFD or similar), many times a second, measuring:
- Latency (delay)
- Jitter (variation in delay)
- Packet loss
SLA policies and steering
| Class | Example SLA | Preferred path |
|---|---|---|
| Voice and video | Under 150 ms, under 30 ms jitter, under 1% loss | MPLS, else best broadband |
| Business SaaS | Under 250 ms, under 2% loss | Local internet breakout |
| Bulk and backups | Best effort | Cheapest link, off-peak |
If the current path stops meeting the SLA, flows move to a better one. SD-WAN reacts to brownouts (a link that is up but performing badly), not just to links going down. Links can also be used active-active, spreading flows across them.
Protecting real-time traffic
- QoS: priority queues for voice on each link, as in any WAN.
- Forward error correction (FEC): sends extra recovery data so lost packets can be rebuilt without retransmission.
- Packet duplication: sends critical packets over two links at once; the first copy to arrive is used.
- TCP optimisation and compression on high-delay links.
Designing good policies
- List the business-critical applications and who uses them.
- Define realistic SLAs per class, based on measured link quality.
- Decide which traffic breaks out locally and which goes to the data centre.
- Start in monitor-only mode, then turn on steering.
- Review reports monthly and adjust.
Real-world example: keeping lessons online
A fictional college runs live remote lessons from three campuses. Each campus has a dedicated internet line and a cheaper broadband line. The network team set SD-WAN policies so that video lessons need under 150 ms latency, under 30 ms jitter and under 1% loss, preferring the dedicated line. Student Wi-Fi and software updates use the broadband line first. During a lesson one afternoon the dedicated line at the smallest campus began dropping packets. Within seconds the SD-WAN moved the lesson's video to the broadband line and lowered the priority of updates there. The lecturer noticed nothing, and the team raised a fault with the provider using the recorded measurements.
Practice
What is a "brownout"?
Which feature sends the same packet over two links?
Which three measurements do SD-WAN probes usually take?
Key takeaways
- SD-WAN identifies applications, then picks a path per application.
- Probes continuously measure latency, jitter and loss on every tunnel.
- SLA policies steer traffic away from failing or degraded links.
- FEC and packet duplication protect voice and video.
- Start in monitor mode, then tune policies with real data.
Original content written by Mohamed Fayas for FM TecHub. © 2026 FM TecHub. You're welcome to share a link to this page; please don't copy the text elsewhere without permission.