Quality of Service (QoS)
Also known as: qos
Quality of Service is a set of network tools that prioritize some traffic over other traffic when a link is congested. It manages bandwidth, delay, jitter, and packet loss so that sensitive applications such as voice and video stay usable.
Without QoS a router or switch forwards packets first come, first served, and a congested link degrades every application equally. That is tolerable for a file download but not for a phone call. QoS lets a network engineer decide which traffic suffers first. It manages four characteristics: bandwidth, delay (latency), jitter (variation in delay), and loss. Voice traffic is the usual benchmark, with common design targets of one-way latency under 150 ms, jitter under 30 ms, and loss under 1%.
Everything begins with classification and marking. Classification identifies traffic using access lists, protocol inspection, or incoming interface. Marking then writes that decision into a header field so downstream devices need not re-inspect the packet. At Layer 2, the class of service (CoS) field inside an 802.1Q trunk header carries three bits with values 0 through 7. At Layer 3, the DSCP field in the IP header carries six bits with values 0 through 63; expedited forwarding (EF, DSCP 46) is the conventional marking for voice, the assured forwarding classes handle prioritized data, and DSCP 0 is best effort. Marking should happen as close to the traffic source as possible, at what is called the trust boundary — typically the access switch port or the IP phone attached to it.
Marked traffic is then acted upon. Queuing places packets into separate queues per class; class-based weighted fair queuing allocates a guaranteed share of bandwidth to each class, and low latency queuing adds a strict-priority queue so voice is always served first. Congestion avoidance with weighted random early detection drops packets from lower-priority classes before a queue fills, which prevents the synchronized TCP slowdown that follows an indiscriminate tail drop.
Two more tools regulate the rate itself. Shaping buffers traffic that exceeds a configured rate and releases it smoothly, adding delay but avoiding drops, and is typically applied outbound. Policing simply drops or remarks traffic above the rate with no buffering, and can be applied in either direction. A common pairing is shaping outbound to match the rate a service provider polices inbound.
The CCNA exam expects you to explain per-hop behaviour for QoS rather than configure it. Know what classification, marking, queuing, congestion avoidance, policing, and shaping each do, the difference between CoS and DSCP and where each lives, the meaning of the trust boundary, and the practical distinction between shaping and policing.
Key takeaways
- QoS prioritizes traffic during congestion by managing bandwidth, delay, jitter, and loss.
- Classification identifies traffic and marking records the decision in the CoS field at Layer 2 or the DSCP field at Layer 3.
- DSCP 46 (expedited forwarding) is the conventional marking for voice traffic.
- Low latency queuing gives voice a strict-priority queue, while WRED drops low-priority packets early to avoid tail drop.
- Shaping buffers excess traffic and adds delay; policing drops or remarks it without buffering.