Quality of Service (QoS) is a set of technologies that allow network administrators to prioritize certain types of traffic over others. It is exclusively used to manage congestion, reduce delay, and prevent Jitter and Packet Loss for business-critical applications.
Traffic Shaping vs Traffic Policing
When engineers want to restrict how much bandwidth an application uses, they use two distinct methods:
- Traffic Shaping: If traffic exceeds the allowed rate, the router buffers (delays) the excess packets in memory and sends them out smoothly over time. Shaping prevents packet loss but increases latency.
- Traffic Policing: If traffic exceeds the allowed rate, the router immediately drops or remarks the excess packets. Policing is strict and causes packet loss, forcing TCP connections to slow down.
The QoS Mechanics: Marking and Queuing
How does a router know a packet is important? Through Traffic Classification and Marking.
When a VoIP phone sends an audio packet, it marks the IP header with a specific DSCP (Differentiated Services Code Point) value (often EF, or Expedited Forwarding). At Layer 2 (Ethernet/VLANs), it might use CoS / 802.1p markings.
When this marked packet reaches a congested router, the router looks at its Queuing and Scheduling configuration. It sees the DSCP EF tag and immediately moves the voice packet to the front of the line (the Priority Queue), forcing a massive file download (tagged as Best Effort) to wait.
A Practical WAN Bottleneck Example
Imagine a branch office with a 100 Mbps WAN link. The current traffic consists of:
- 2 Mbps of Voice traffic
- 10 Mbps of Video Meetings (Google Meet / Teams)
- 20 Mbps of Business CRM traffic
- 80 Mbps of Server Backup traffic
Total: 112 Mbps. The link is congested. Without QoS, the router drops packets randomly. A dropped voice packet causes terrible call quality. With QoS configured, the router places the Voice and Video into a strict priority queue, ensuring they never drop a packet. The router forces the Server Backup traffic to absorb 100% of the packet loss.
Warning: Marking traffic is useless unless every device along the path (especially the bottleneck device) is configured to actually honor those markings. If you mark traffic but your ISP strips the tags, the QoS is destroyed.
Common QoS Troubleshooting
- Wrong Classification: The firewall is identifying Teams video as "web browsing" and placing it in the slow queue.
- DSCP Rewritten: A switch in the middle of the network is untrusting and resetting all DSCP values to 0 (Best Effort).
- Packet Loss Mistaken for QoS Failure: The QoS policy is working perfectly, but the ISP link has a damaged fiber cable dropping 10% of all traffic physically.
- Insufficient Bandwidth: If you have 50 Mbps of Voice traffic trying to cross a 20 Mbps link, QoS cannot save you. You need more bandwidth.
Best Practices
- Classify business-critical traffic carefully. Do not prioritize everything, or priority loses its meaning.
- Measure Latency and Throughput before and after applying QoS policies.
- Apply QoS policies precisely at the bottleneck (usually the WAN edge router).