QoS for Voice, Video and Business Traffic Playbook

QoS does not create bandwidth

QoS decides how traffic is treated when resources are contested. If a WAN link is continuously undersized, QoS can protect important traffic but it cannot manufacture extra capacity.

The trust boundary

The important question is not only "what DSCP value should voice use?" but "which device are you willing to trust to mark traffic correctly?" Cisco Catalyst 9000 supports conditional trust for recognised device types and can remark untrusted traffic. This is useful on ports where a phone is trusted but a downstream PC should not be able to mark ordinary downloads as priority traffic.

Classification and marking

Traffic can be classified by DSCP, VLAN, ACL, protocol or other fields depending on platform. Marking assigns a value that downstream devices can use. A complete design defines where classification happens and where markings are trusted.

Queuing

At a congested egress interface, different traffic classes compete for queue resources. Low-latency traffic such as voice may receive priority treatment, while ordinary bulk traffic waits or drops according to policy.

Policing

Policing enforces a rate and can drop or remark excess traffic. Cisco's Catalyst 9000 QoS documentation distinguishes the configured rate from burst allowance. Use policing carefully because too little burst allowance can damage applications even when the average rate appears acceptable.

Vendor example: AutoQoS

interface GigabitEthernet1/0/12
 auto qos voip cisco-phone

Cisco documents AutoQoS templates for voice, telepresence and IP-camera scenarios. Treat AutoQoS as a starting template for supported platforms, then inspect the generated policy and verify it matches the real application.

How to commission QoS

1. Measure whether the path actually experiences congestion.
2. Identify application traffic and required treatment.
3. Define the trust boundary.
4. Apply classification/marking policy.
5. Generate representative traffic.
6. Verify queue counters and drops during congestion.
7. Confirm ordinary traffic still has sufficient service.

What works

• End-to-end policy consistency.
• Trusting markings only from controlled devices or network boundaries.
• Verifying queue counters under real congestion.

What does not work

• Enabling QoS and assuming voice is now protected everywhere.
• Trusting DSCP from every desktop.
• Trying to fix a permanently saturated link only with priority queues.