Quality of Service · WAN egress
Generate a clean MQC QoS service-policy for voice and video over WAN. Outputs class-map, policy-map, LLQ priority percentages, DSCP markings, and the interface bindings so you can paste into IOS-XE and go.
The number one cause of "one-way audio" and "robot voice" on Cisco networks isn't codecs or trunks — it's a missing or broken WAN egress QoS policy. This tool builds Cisco's recommended 8-class MQC policy (with sensible simplifications for small sites), so voice gets LLQ, video gets bandwidth, signaling is protected, and everything else plays nicely.
LLQ (Low Latency Queue) gives voice unconditional priority up to its policed limit. Above that limit, excess voice packets are dropped — which is by design, because you never want voice starving other traffic if someone misconfigures a codec. The number the tool picks is ceil(calls × 100 kbps × 1.25) to cover overhead.
If the voice reservation exceeds 33% of WAN bandwidth, you're either over-provisioning voice or under-provisioning WAN. The tool will warn. Video uses CBWFQ bandwidth (not priority) so it shares gracefully with data under congestion. Signaling (CS3) gets a small guaranteed slice — usually 2-5%.
Cisco's enterprise baseline, circa IOS-XE 17.x, is an 8-class model: voice (EF), interactive video (AF4), streaming video (AF3), call signaling (CS3), network control (CS6), transactional data (AF2), bulk data (AF1), and best-effort. For most branch WANs, 4 classes are enough: voice, video, signaling, everything else. The tool supports both.
If your endpoints mark correctly (Cisco IP phones mark EF for RTP and CS3 for SCCP/SIP signaling; Webex App marks DSCP via Group Policy; Teams marks EF when the tenant policy pushes it), the policy can trust on ingress and apply on egress. Otherwise, re-mark at the WAN edge using an access-list-based class-map.
Metro Ethernet, DIA, and SD-WAN underlays are almost always subrate — your interface is GigabitEthernet but your contract is 50 Mbps. Without shape average, the router fires traffic at line rate and the carrier polices / drops. Always shape to the contracted rate (minus 5% for headroom), and nest the QoS policy inside the shaper.
Yes. Microbursts can fill a 1 Gbps pipe for 50 ms, which is enough to delay voice past the 150 ms one-way threshold. QoS costs nothing when idle and saves you when you need it.
Same thing. EF is the PHB name, 46 is the decimal. Cisco docs use both interchangeably.
Cisco Viptela (now Catalyst SD-WAN) uses template-based QoS. The underlying class structure is the same — this tool's output translates directly; you paste the class-maps into the device template and apply the policy-map as a localized data policy.
Percent is safer for sub-rate circuits because the shaper is the source of truth; priority with a fixed kbps can mismatch. The tool uses percent when shaping is enabled.
Voice LLQ = 20 × 100 × 1.25 = 2500 kbps → 5% of 50 Mbps. Video = 30% (15 Mbps). Signaling = 5%. Best-effort = 60%. Verdict: OK, voice reservation under the 33% guardrail.
Voice LLQ = 6 × 100 × 1.25 = 750 kbps → 8% of 10 Mbps. Signaling = 5%. Best-effort = 87%. 4-class model recommended because this site has no video.
Voice LLQ = 100 × 100 × 1.25 = 12.5 Mbps → ~1.3% of 1 Gbps. Video = 20% (200 Mbps). Signaling = 3%. Loads of headroom for data. Often this campus doesn't need shaping — drop the shaper and apply the policy directly on the interface.
Drop your email and we'll send your generated QoS policy plus a MQC config walkthrough as PDF.