Packet Classification & QoS Marking — Architecture & Mechanisms

Overview

Packet classification and QoS marking determine how packets are identified, categorized, and labeled so that downstream components—schedulers, shapers, policers, and congestion‑control mechanisms—can apply differentiated treatment. These mechanisms form the foundation of Quality of Service (QoS), enabling networks to prioritize latency‑sensitive traffic, enforce bandwidth guarantees, and isolate flows.

Classification assigns packets to traffic classes or flows.
Marking encodes QoS information into packet headers for use by intermediate nodes.

Together, they define how traffic is recognized and how it should be treated across the network.

Classification Objectives

Packet classification is designed to achieve several goals:

• Identify flows for scheduling, shaping, and policing
• Apply QoS policies based on application requirements
• Enable differentiated services (voice, video, best‑effort, control traffic)
• Support security and access control
• Enable traffic engineering through class‑based routing or prioritization

Classification is the first step in the QoS pipeline.

Architectural Principles

Multi‑Field Classification

Packets are classified by examining multiple header fields, such as:

• source/destination IP
• source/destination port
• protocol type
• VLAN ID
• DSCP/ToS bits
• MPLS labels
• flow identifiers
• application signatures (in advanced systems)

Multi‑field classification enables fine‑grained control.

Hierarchical Classification

Classification may occur at multiple levels:

• per‑packet
• per‑flow
• per‑class
• per‑tenant
• per‑service

Hierarchical classification supports scalable QoS in large networks.

Stateless vs Stateful Classification

• Stateless classification examines only the current packet.
• Stateful classification tracks flow state (e.g., TCP connection, session metadata).

Stateful classification enables more accurate QoS decisions but requires more memory and logic.

Classification Mechanisms

Exact Match Classification

Packets are matched against exact header values.

Use cases: MAC tables, ARP caches, simple ACLs.

Longest Prefix Match (LPM)

Used for IP routing and class‑based forwarding.

Use cases: routing tables, class‑based routing.

Ternary Content‑Addressable Memory (TCAM)

TCAM enables matching with wildcards and masks.

Pros: extremely fast
Cons: high power, limited size

Used in high‑performance switches and routers.

Hash‑Based Classification

Flows are hashed into buckets.

Pros: scalable
Cons: collisions may cause misclassification

Used in load balancers and fair‑queuing systems.

Deep Packet Inspection (DPI)

Examines payload to identify applications.

Pros: accurate application classification
Cons: expensive, privacy concerns, encrypted traffic limits effectiveness

Used in enterprise and carrier networks.

QoS Marking

Purpose of Marking

Marking encodes QoS intent into packet headers so that downstream nodes can apply:

• priority
• bandwidth guarantees
• shaping
• policing
• congestion handling

Marking ensures consistent treatment across the network.

Marking Fields in Common Protocols

Ethernet (802.1Q)

• PCP (Priority Code Point) — 3 bits
  Defines 8 priority levels (0–7).
  Used in VLAN‑tagged Ethernet frames.

IP (IPv4/IPv6)

• DSCP (Differentiated Services Code Point) — 6 bits
  Defines traffic classes such as EF (Expedited Forwarding), AF (Assured Forwarding), and BE (Best Effort).
• ECN (Explicit Congestion Notification) — 2 bits
  Used for congestion signaling without packet loss.

MPLS

• Traffic Class (TC) — 3 bits
  Used for QoS and ECN in MPLS networks.

Wi‑Fi (802.11e/WMM)

• Access Categories (AC)
  Maps traffic into four priority classes:

• Voice
• Video
• Best Effort
• Background

QoS Models

Best Effort

No guarantees; all packets treated equally.

Differentiated Services (DiffServ)

Packets are marked with DSCP values; routers apply per‑hop behaviors (PHBs).

Pros: scalable
Cons: coarse‑grained

Integrated Services (IntServ)

Uses per‑flow reservations (RSVP).

Pros: strong guarantees
Cons: not scalable for large networks

Class‑Based QoS

Traffic is grouped into classes with:

• priority
• bandwidth guarantees
• shaping policies

Used in data centers and carrier networks.

Classification in Modern Systems

Data Center Networks

Classification supports:

• tenant isolation
• priority for latency‑sensitive flows
• ECN‑based congestion control (DCTCP)
• traffic engineering

DSCP and VLAN PCP are commonly used.

Carrier and ISP Networks

Classification enables:

• service differentiation
• subscriber QoS
• traffic engineering
• SLAs

MPLS TC and DiffServ are widely deployed.

Wireless Networks

Classification interacts with:

• airtime fairness
• channel‑aware scheduling
• QoS class identifiers (QCIs) in LTE/5G

Wireless systems rely heavily on priority‑based classification.

High‑Speed Interconnects

PCIe, CXL, and NoCs classify traffic into:

• control
• data
• completion
• coherence classes

Scheduling and flow control depend on class identifiers.

Performance Considerations

Latency

Higher‑priority classes receive lower latency.
Misclassification can cause jitter or delay.

Throughput

Weighted classes ensure bandwidth guarantees.
Incorrect marking may cause unfairness.

Scalability

TCAM‑based classification is fast but expensive.
Hash‑based classification scales better but is less precise.

Security

Classification interacts with ACLs and firewall rules.
Incorrect classification may expose attack surfaces.

Comparison of Classification Approaches

Approach	Precision	Scalability	Complexity	Typical Use
Exact Match	High	Medium	Low	MAC tables
LPM	Medium	High	Medium	Routing
TCAM	Very High	Low	High	QoS, ACLs
Hash‑Based	Medium	Very High	Low	Load balancing
DPI	Very High	Low	Very Low	Application QoS

Design Tradeoffs

• Precision vs scalability — TCAM is precise but costly; hashing scales but may collide.
• Latency vs complexity — deeper classification increases latency.
• QoS guarantees vs fairness — strict priority may starve lower classes.
• Security vs performance — DPI improves security but increases overhead.
• Consistency vs flexibility — marking must be consistent across the network.

Related Pages

• Packet Scheduling — Architecture & Algorithms
• Traffic Shaping & Policing — Architecture & Algorithms
• Queue Management & Congestion Control — Architecture & Algorithms
• Protocol Flow Control — Architecture & Mechanisms
• Packet Fragmentation & Reassembly — Architecture & Design Considerations

Summary

Packet classification and QoS marking define how packets are identified and labeled for differentiated treatment across the network. Through multi‑field classification, DSCP/PCP marking, and class‑based QoS models, modern systems enforce fairness, prioritize latency‑sensitive traffic, and maintain predictable performance across diverse environments.