Table of Contents Link to heading

• Latency
• Quality of Service (QoS) Models
  • Best effort
  • Integrated Services (IntServ)
  • Differentiated Services (DiffServ)
• Modular QoS CLI (MQC)
  • class-map
  • policy-map
  • service-policy
  • Core MQC Actions
    • Priority Queue (LLQ – Low Latency Queue)
    • Bandwidth Guarantee (Class-based weighted fair queuing (CBWFQ))
    • Policing (Traffic Rate Limiting)
    • Shaping (Traffic Smoothing)
    • Verification
• Classification
• Marking
  • Layer 2 Marking - Class of Service (CoS)
  • Layer 3 Marking
    • IP Precedence (IPP)
    • DiffServ Code Point (DSCP)
      • Per-Hop Behaviours (PHB)
        • Class Selector (CS)
        • Default Forwarding (DF)
        • Expedited Forwarding (EF)
        • Assured Forwarding (AF)
    • Trust Boundary
• Policing
• Shaping
• Congestion Management
  • FIFO (First In, First Out)
  • Round Robin (RR)
  • Weighted Round Robin (WRR)
  • Weighted Fair Queuing (WFQ)
  • Priority Queuing (PQ)
  • Custom Queuing (CQ)
  • CBWFQ (Class-Based WFQ)
  • LLQ (Low Latency Queuing)
• Congestion Avoidance
  • Tail Drop
  • RED (Random Early Detection)
  • WRED (Weighted RED)

Latency Link to heading

It is recommended that a network latency should not exceed 400 ms, and for real-time traffic, network latency should be less than 150 ms.

Causes of network latency include:

• Propagation delay (fixed) is the time it takes for a packet to travel from the source to a destination at the speed of light over a medium such as fiber-optic cables or copper wires.
• Serialisation delay (fixed) is the time it takes to place all the bits of a packet onto a link
• Processing delay (fixed) is the fixed amount of time it takes for a networking device to take the packet from an input interface and place the packet onto the output queue of the output interface.
• Delay variation (variable) or jitter describes how packets arrive at the destination at varying intervals.
  • Low jitter = smooth voice/video
  • High jitter = choppy audio, frozen video, poor user experience

Quality of Service (QoS) Models Link to heading

Best effort Link to heading

Integrated Services (IntServ) Link to heading

IntServ uses Resource Reservation Protocol (RSVP) to reserve resources throughout a network for a specific application and to provide call admission control (CAC) to guarantee that no other IP traffic can use the reserved bandwidth.

To be able to provide end-to-end QoS, all nodes, including the endpoints running the applications, need to support, build, and maintain RSVP path state for every single flow.

• Created for real-time applications
• Unused reserved bandwidth is wasted
• Highest guarantee of QoS but resource-intensive
• Precise but not scalable

Differentiated Services (DiffServ) Link to heading

DiffServ divides IP traffic into classes and marks it based on business requirements.

As IP traffic traverses a network, each network device identifies the packet class by its marking and services the packets according to this class.

Modular QoS CLI (MQC) Link to heading

It is a 3-step framework:

1. class-map     → Identify traffic
2. policy-map    → Decide what to do
3. service-policy→ Apply to interface

class-map Link to heading

Used to match traffic.

Common match methods:

All traffic that is not matched by any of the user-defined traffic classes is referred to as unclassified traffic, and is treated by the default class called class-default.

Unclassified traffic, if not applied any QoS actions, is best-effort traffic.

ip access-list extended VOICE-TRAFFIC
 10 permit udp any any range 16384 32767
 20 permit udp any range 16384 32767 any

class-map match-all VOIP-TELEPHONY
 match access-group name VOICE-TRAFFIC
 match dscp ef

class-map match-any P2P
 match protocol bittorrent
 match protocol soulseek

policy-map Link to heading

Used to define action per class

Common QoS Actions:

policy-map QUEUING
 class VOIP-TELEPHONY
 priority level 1 percent 30
 class P2P
 bandwidth percent 10
 class class-default

service-policy Link to heading

Attach policy to interface.

interface GigabitEthernet1
 service-policy output QUEUING

Core MQC Actions Link to heading

Priority Queue (LLQ – Low Latency Queue) Link to heading

Used for:

• Voice
• Video conferencing
• Real-time applications

class VOICE
 priority percent 30

📌 Key rules:

• Strict priority
• Policed automatically
• Prevents starvation

Bandwidth Guarantee (Class-based weighted fair queuing (CBWFQ)) Link to heading

class DATA
 bandwidth percent 40

Guarantees minimum bandwidth under congestion.

Policing (Traffic Rate Limiting) Link to heading

Used to drop or remark excess traffic

class BULK
 police 10000000 conform-action transmit exceed-action drop

10 Mbps policing.

Shaping (Traffic Smoothing) Link to heading

Used mainly on WAN links

policy-map SHAPE-10M
 class class-default
  shape average 10000000

Verification Link to heading

show policy-map interface gi0/1
show class-map
show policy-map

Classification Link to heading

Packet classification uses traffic descriptors to categorise an IP packet within a specific class.

Packet classification should take place at the network edge, as close to the source of the traffic as possible.

The following traffic descriptors are typically used for classification:

• Internal: QoS groups (locally significant to a router)
• Layer 1: Physical interface, subinterface, or port
• Layer 2: MAC address and 802.1Q/p class of service (CoS) bits
• Layer 2.5: MPLS experimental (EXP) bits
• Layer 3: Differentiated Services Code Points (DSCP), IP Precedence (IPP), and source/destination IP address
• Layer 4: TCP or UDP ports
• Layer 7: Next-Generation Network-Based Application Recognition (NBAR2)

Marking Link to heading

Packet marking uses a traffic descriptor to change a field within a packet or a frame header.

The following traffic descriptors are used for marking traffic:

• Internal: QoS groups
• Layer 2: 802.1Q/p class of service (CoS) bits
• Layer 2.5: MPLS experimental (EXP) bits
• Layer 3: Differentiated Services Code Points (DSCP) and IP Precedence (IPP)

Layer 2 Marking - Class of Service (CoS) Link to heading

• Found in the 802.1Q VLAN field (32 bits)
• A 3-bit 802.1p field named Priority Code Point (PCP)
• Drawback: frames lose their CoS markings when traversing a non-802.1Q link or a Layer 3 network.

Layer 3 Marking Link to heading

• Found in the DiffServ (DS) field (8 bits)
• Legacy name is Type of Service (ToS) field
• Preserved end-to-end and mitigates Layer 2 marking’s drawback.

IP Precedence (IPP) Link to heading

• A 3-bit field
• Replaced by DSCP

DiffServ Code Point (DSCP) Link to heading

• A 6-bit field
• Routers apply Per-Hop Behaviours (PHB) at every hop based on DSCP.

Per-Hop Behaviours (PHB) Link to heading

Class Selector (CS) Link to heading

CS values map to old IP Precedence

DSCP = CSn = 8n

Default Forwarding (DF) Link to heading

Used for best-effort service, equal to CS0.

Expedited Forwarding (EF) Link to heading

EF packets are placed in a strict priority queue with minimal delay, jitter, loss.

Routers forward EF traffic immediately, ahead of other classes.

Fixed DSCP: 46

Used for:

• IP Telephony
• Voice RTP
• Critical interactive traffic

Assured Forwarding (AF) Link to heading

AF ensures that traffic receives better-than-best-effort service but does not guarantee absolute delivery.

AF format:

AFxy
x = class priority (1–4)
y = drop precedence (1–3)

DSCP = 8x + 2y

Trust Boundary Link to heading

The point where the network stops trusting incoming DSCP markings.

• Trust IP phones (DSCP 46)
• Do NOT trust PCs (remark PC -> BE or CS0)
• Used only on phone ports / uplinks

interface g0/1
 mls qos trust device cisco-phone
 mls qos trust cos

interface g0/2
 mls qos cos override

Policing Link to heading

• Purpose: Enforce traffic rate limits by dropping or remarking excess traffic with a lower priority.
• Used for: ingress and egress traffic
• Example: Limit guest network traffic to 2 Mbps, and drop excess packets if the limit is exceeded.

Shaping Link to heading

• Purpose: Smooth traffic flow by buffering and delaying excess traffic.
• Used for: egress traffic
• Example: Shape video traffic to 5 Mbps, buffering excess packets to avoid bursts.

Congestion Management Link to heading

FIFO (First In, First Out) Link to heading

• One queue
• First packet in = first out
• No classification, no priority
• Drops packets during congestion (tail drop)
• Default when QoS is not configured

Round Robin (RR) Link to heading

• Multiple queues
• Each queue sends 1 packet per turn
• Equal turns, not equal bandwidth
• No priority

Weighted Round Robin (WRR) Link to heading

• Like RR but with weights
• Higher weight = more packets per round
• Still affected by packet size

Weighted Fair Queuing (WFQ) Link to heading

• Automatically classifies flows
• Each flow gets a queue
• Small/interactive flows preferred
• Used on low-speed links (legacy behaviour)

Priority Queuing (PQ) Link to heading

• Multiple priority levels (High → Low)
• Always serves highest queue first
• Can cause starvation

Custom Queuing (CQ) Link to heading

• Admin defines:
  • Number of queues
  • Bytes per queue per round
• Prevents starvation
• Legacy feature

CBWFQ (Class-Based WFQ) Link to heading

• You define traffic classes
• Assign bandwidth per class
• No strict priority by default
• Main enterprise method

LLQ (Low Latency Queuing) Link to heading

• CBWFQ + strict priority queue (PQ)

Congestion Avoidance Link to heading

Tail Drop Link to heading

• Default method
• Drops packets only when queue is full
• Can cause global synchronisation (all TCP flows slow down together)
• Reactive (too late)

RED (Random Early Detection) Link to heading

• Drops packets before queue is full
• Based on average queue depth
• Drops randomly
• Reduces global synchronisation
• Not QoS-aware

WRED (Weighted RED) Link to heading

• RED + DSCP/IP Precedence aware
• Lower-priority traffic dropped earlier
• Higher-priority traffic dropped later
• Used with CBWFQ / DiffServ