Quality of Service (QoS)

Overview

QoS prioritizes delay-sensitive traffic (voice, video) over bulk data through classification, marking, queuing, and shaping. CCNA focuses on the classify → mark → queue pipeline and trust boundaries at the access layer.

Why QoS exists

Without QoS, all traffic competes equally for bandwidth. Voice and video are delay-, jitter-, and loss-sensitive; large data transfers are greedy and delay-tolerant. QoS provides managed unfairness — priority to critical applications.

One-way requirements (voice/video)

These are one-way — each direction must meet targets independently.

Traffic types

Interactive data (Telnet) is more delay-sensitive than bulk transfer (FTP download).

Congestion

When output rate exceeds link capacity, packets queue. Queues cause delay; changing queue depth causes jitter; full queues cause drops. Voice calls degrade quickly under congestion without QoS.

The QoS pipeline

1. Classification — identify traffic (ACL, NBAR, CoS, DSCP)
2. Marking — label packets (CoS at L2, DSCP at L3)
3. Queuing / scheduling — prioritize dequeue order
4. Shaping / policing — rate-limit traffic

Classify and mark as close to the source as possible (edge of network). Every hop uses the marking to apply consistent treatment.

Classification and marking

Layer 2 — CoS (Class of Service)

802.1Q trunk headers include a 3-bit CoS field (values 0–7):

CoS only exists on tagged frames — requires a trunk between phone and switch (or switch and router).

Layer 3 — DSCP

The IPv4/IPv6 DSCP field uses 6 bits (64 values) in the former ToS byte:

Assured Forwarding (AF) classes use the first 3 bits for class (AF1–AF4) and next 2 bits for drop probability — higher drop probability = more likely to be discarded under congestion.

Other classification methods

Trust boundaries

The trust boundary is where the network decides whether to accept or rewrite QoS markings from an endpoint.

Mark at the edge; classify on every device along the path using the marking.

MQC — Modular QoS CLI

Cisco QoS uses three building blocks:

class-map match-all VOICE
match access-group name VOICE-ACL
!
policy-map WAN-EDGE
class VOICE
set dscp ef
class class-default
fair-queue
!
interface GigabitEthernet0/0
service-policy output WAN-EDGE

Full MQC policy config is awareness-level for CCNA — focus on concepts and markings.

Queuing mechanisms

Queuing activates during congestion (output queue full). Scheduling decides dequeue order.

LLQ (Low Latency Queuing) gives voice a strict priority queue with policing — voice goes first, but capped so it cannot starve other traffic.

WRED (awareness)

Weighted Random Early Detection randomly drops packets before the queue is full to signal TCP senders to slow down — avoids tail-drop synchronization. Typically applied to TCP data, not UDP voice.

Shaping vs policing

Both measure traffic rate against a configured limit:

Tri-color policing: conform (transmit) → exceed (re-mark) → violate (drop).

AutoQoS (awareness)

Cisco AutoQoS macros generate MQC policies for common designs (voice + data). Useful in labs; understand what it accomplishes, not every macro keyword.

End-to-end voice QoS checklist

1. Phone marks voice EF / CoS 5 and signaling CS3 / CoS 3
2. Access switch trusts phone CoS on voice VLAN port
3. Queuing (LLQ) on WAN edge gives priority to EF
4. Bandwidth guarantees for signaling and video classes
5. Avoid congestion where possible; QoS mitigates, does not create bandwidth

See Voice VLAN topic for phone/switch integration.

Exam checklist