FamousTec

WAN · MPLS CE queuing

MPLS CE Queue Plan

Build a full MPLS CE egress queue plan aligned to your carrier's CoS model. Outputs class-map, policy-map, WRED thresholds, shaper, and the DSCP-to-EXP table-map you must agree with the carrier.

MPLS CE routers carry every branch's voice, video, and data. One misplaced class or a mis-sized priority queue and your whole voice quality story falls apart across 80 sites.

How to interpret the output

The tool produces a parent shaper = CIR and a child policy with LLQ priority for voice, CBWFQ for video, signaling, transactional, and best-effort. WRED is applied to best-effort to prevent TCP global synchronization. Voice is priority-percent so it floats with the shaper.

If voice + video exceeds 50% of CIR, the tool flags over-commit: under burst, business traffic will starve. Either raise CIR or reduce voice/video allocations.

Deep dive — CE queuing anatomy

Parent shaper sets the total rate. Child policy is queued on top. On IOS-XE Catalyst 8000 family, HQF (Hierarchical QoS Framework) handles this in hardware. Without a parent shaper, the child policy would fire at the physical port rate and the PE router would police your CIR, dropping whatever exceeds — including voice.

Burst tolerance

Carriers allow burst up to some Bc/Be value. You want to sit at CIR or slightly under, so bursts of best-effort never touch voice LLQ. The tool uses CIR × 0.95 as shape rate to preserve a 5% safety margin.

Why trust this tool

Built after a decade of inheriting broken MPLS CE configs where someone had shaped at port speed and wondered why calls were choppy at 3 PM.

FAQ

Priority percent or priority kbps?

On a shaper, always percent — it floats with the shaped rate. On a physical interface without shaping, kbps is fine.

Do I need WRED on voice?

No. Voice is UDP, WRED would just drop packets that are going to be dropped anyway by LLQ policing. WRED goes on TCP classes (best-effort, transactional).

What about H-QoS for multi-VLAN CEs?

Nest per-VLAN policies inside a parent shaper for total CIR. IOS-XE supports 3-level hierarchies on Catalyst 8000 family.

Worked examples

200 Mbps Lumen 4-class, branch HQ

Voice 15% (30 Mbps) = ~300 concurrent calls. Video 25% (50 Mbps). Signaling 5%. Best-effort 55%. Shape 190 Mbps.

1 Gbps AT&T 4-class, regional DC

Voice 5% (50 Mbps). Video 25% (250 Mbps). Critical/Business 20%. BE 50%. Shape 950 Mbps.

100 Mbps Verizon 6-class, small branch

Map 6 DSCP classes into VZ's EF/AF41/AF31/AF21/AF11/BE equivalents. Voice 10%. Video 20%. Shape 95 Mbps.

References

Want this report by email?