In the field of network performance management, Quality of Service (QoS) technology regulates the priority and resource allocation of data flow like a traffic signal system. Its core value is to balance the traffic competition during network congestion and ensure the quality of service for critical services. In this paper, we systematically elaborate the core logic and engineering practice of QoS from the three dimensions of technical realization, scenario adaptation, and strategy selection.
1. The technical nature and realization mechanism of Quality of Service
1. Definition and objectives
QoS is a network resource management technology that realizes the following goals by classifying, marking and scheduling data streams:
Bandwidth guarantee: Reserve enough bandwidth for real-time services such as video conferencing and VoIP.
Delay control: Reduce end-to-end delay in scenarios such as gaming and industrial control
Packet loss suppression: Reduce the probability of packet loss during critical data transmission
2. Core Technology Components
The Quality of Service technology stack contains four core modules:
Classification: Identify traffic types based on quintuple (source/destination IP, port, protocol) or DSCP marking.
Marking: Define priorities using DiffServ’s DSCP (6bit) or IEEE 802.1p (3bit) fields.
Queuing: scheduling packets using algorithms such as WFQ (Weighted Fair Queuing), CBWFQ (Class Based Weighted Fair Queuing), and so on.
Traffic shaping (Shaping): through the token bucket algorithm to limit burst traffic, to avoid link congestion
Take the typical configuration of a Cisco device as an example:
class-map match-any VOICE
match dscp ef
policy-map QOS-POLICY
class VOICE
priority percent 20
class VIDEO
bandwidth percent 30
class default
fair-queue
This policy allocates 20% of strictly prioritized bandwidth for voice traffic, guarantees 30% bandwidth for video traffic, and schedules the rest of the traffic fairly.
2.Typical Application Scenarios and Policy Selection of Quality of Service
1. Corporate Office Networks
Demand characteristics: guarantee the priority of video conferencing and ERP system, and limit the bandwidth occupied by P2P downloads.
Configuration scenario:
Mark SIP/RTP protocol as EF (accelerated forwarding)
Implement speed limit on HTTP download traffic (Police 10Mbps)
Enable LLQ (Low Latency Queuing) to ensure voice packets are prioritized for transmission
2. Industrial Internet of Things
Requirements: low latency (≤10ms) for PLC control commands, high reliability of sensor data.
Technology Implementation:
Use 802.1p priority tagging for OPC UA real-time data
Configuration of separate hardware queues for PROFINET IRT traffic
Deployment of TSN (Time Sensitive Networking) to enhance clock synchronization accuracy
3. Cloud computing and virtualization
Challenge: Resource isolation when multi-tenants share physical links
Solution:
Apply SR-IOV+DCB (Data Center Bridging) technology at the vSwitch layer.
Dynamically adjust QoS policies based on SDN controller (e.g. OpenFlow Meter table)
Enabling CNI QoS plug-ins for container networks (e.g. Kubernetes Bandwidth API)
3. Decision Modeling for Enabling and Disabling Quality of Service
1. Recommended scenarios to enable Quality of Service
Frequent network congestion: When the link utilization rate continuously exceeds 70%, QoS needs to be enabled to avoid damage to critical services.
Large differences in business sensitivity: there are real-time audio and video, industrial control and other high-priority traffic mixed with ordinary data transmission.
Multi-tenant resource sharing: cloud computing or enterprise branches need to protect the SLAs of different departments/customers.
2. Scenarios to consider disabling Quality of Service
Extremely abundant network bandwidth: for example, the actual load of 10Gbps link is less than 10% for a long time.
Device performance bottleneck: Enabling QoS on low-end routers may lead to CPU overload (e.g., 50% drop in forwarding performance).
Protocol compatibility issues: Some old devices do not support DSCP marking, resulting in policy failure.
3. Compromise solution: selective enablement
Interface-level control: Enable QoS only on bottleneck links such as WAN egress, and keep the default scheduling within the LAN.
Time policy: Enable QoS during peak hours (9:00-18:00) and disable it at night to reduce equipment load.
Business-level granularity: Enable priority protection only for critical applications (e.g., SAP, Zoom).
4. Engineering Practice Points for Quality of Service Configuration
1. Accuracy Guarantee for Traffic Classification
Deep Packet Inspection (DPI): Identify the actual application type of encrypted traffic (e.g., determine web services by TLS SNI field)
Machine learning assistance: automatic generation of classification rules based on historical traffic characteristics, adapting to new types of applications (such as meta-universe traffic)
2. Optimization and adjustment of queuing parameters
Buffer depth: dynamically calculated based on link RTT (round-trip delay) to avoid buffer bloat (Bufferbloat)
WRED (Weighted Random Early Detection): implement active packet loss in TCP traffic to prevent global synchronization problems
3. Cross-device policy consistency
End-to-end token retention: Configure Trust Boundary to ensure DSCP tokenization by switches and routers.
Centralized SDN control: Uniformly issue QoS policies through OpenDaylight and other controllers to eliminate configuration differences between devices.
5. Typical Configuration Mistakes and Troubleshooting Guidelines
1. Common Configuration Mistakes
Priority flooding: too much traffic is marked as high priority, losing the significance of scheduling.
Excessive rate limitation: Police value is set lower than the business requirements, resulting in legitimate traffic being discarded.
Hardware queuing is not enabled: only software QoS is configured, which cannot meet the low latency requirements.
2. Performance Problem Troubleshooting Path
Baseline test: Measure network performance after disabling QoS to confirm whether QoS is causing bottlenecks.
Queue monitoring: View queue discard statistics through the show policy-map interface.
Marking verification: Use Wireshark to capture packets to check whether DSCP/802.1p marking is correct.
Conclusion: The Value Nature of Quality of Service and Technology Evolution
QoS is not just a bandwidth allocation tool, but a value scheduling system for network resources. Its ultimate goal is to maximize overall business performance through differentiated management. With the development of AI technology and Intent-Based Networking, the next generation of QoS will have the following characteristics:
Dynamic policy generation: Automatically derive QoS parameters based on business objectives.
Cross-domain collaboration: Realize end-to-end QoS from endpoint to cloud
Enhanced Perception: Integration of application layer semantic understanding to improve classification accuracy
Enterprises need to find a balance between “fine-grained control” and “management complexity” based on business requirements, network status quo and technical capabilities. In most modern network environments, selectively enabling QoS is still a necessary option to ensure critical business experience.
