QUALITY OF SERVICE

CLASS 1
INTRODUCTION
About the instructor
• Name: Guilherme IECKER RICARDO
• Office : F-402
• Email : [email protected]
• Web page : http://irit.fr/~Guilherme.Ricardo
COURSE
PRESENTATION
Course Material
• No textbook !
• Moodle : N8EN18A - Qualité de Service
• Course slides
• TP and TD handouts
• Additional resources (e.g., papers, software)
Course Organization
Class Data Time Topic
CM1 24/01/2024 16h15 - 18h00 Networking, traffic characteristics and fairness
CM2 08/02/2024 08h00 - 10h00 Classification, Marking, and Congestion avoidance
CM3 13/02/2024 14h00 - 15h45 Policing, Shaping, and Scheduling
CM4 14/02/2024 14h00 - 15h45 Traffic Engineering: MPLS and RSVP
R: 08h00 - 09h45
TD 16/02/2024 -
A: 16h15 - 18h00
R: 01/03/2024 14h00 - 15h45
TP1 TCP Flows and Congestion Control
A: 01/03/2024 16h15 - 18h00
CM5 05/03/2024 10h15 - 12h00 QoS in modern mobile networks
R: 06/03/2024 14h00 - 15h45
TP2 Packet Processing and Queuing
A: 08/03/2024 08h00 - 09h45
Éxamen 22/03/2024 10h15 - 12h00 -
Before we start...

Objectives:
• Review of fundamental network
operations and performance
indicators
• Different traffic characteristics
• Definition of QoS
RECAP OF L3
NETWORKING
The Internet Networking Model
Layered Abstraction of End-to-End Communication

Source: Kurose, Computer Networking. 7th ed.

Network Layer
Routing vs. Forwarding

Source: Kurose, Computer Networking. 7th ed.

Packet Switching
Store-and-Forward Transmission

Source: Kurose, Computer Networking. 7th ed.

Packet Switching
Queuing

Source: Kurose, Computer Networking. 7th ed.

Packet Switching
Transmission Stages

Source: Kurose, Computer Networking. 7th ed.

Packet Switching
Router Architecture

Source: Kurose, Computer Networking. 7th ed.

Packet Switching
Router Architecture

Source: Kurose, Computer Networking. 7th ed.

Packet Switching
Router Architecture

Source: Kurose, Computer Networking. 7th ed.

Packet Switching
Router Architecture

Source: Kurose, Computer Networking. 7th ed.

Managing Output Queues
Queuing Theory Representation

μ
λ
Managing Output Queues Parameters:
• Service rate (output link capacity)
Parameters and Scheduling • Number of queues
• Queue(s) size
• Arrival rate

μ
λ

Scheduling Disciplines: Active Queue Management (AQM)

• First-In-First-Out (FIFO) • Dropping
• Round Robin • Marking
• Weighted Fair Queuing (WFQ) • Random Early Detection (RED)
• etc. • etc.
Network Service Models
General Network Services

Guaranteed Delivery • Every generated packet is delivered to the destination host, eventually

Delay-bounded • Every generated packet is delivered to the destination host

Guaranteed Delivery • The packet arrives within a given delay tolerance

In-Order Delivery • Packets arrive at the destination in the order that they were sent

• Every generated packet is delivered to the destination host, if it is sent

Minimal-Rate Delivery
at a rate larger than a given bound

• Datagram’s payload is encrypted at the source and only decrypted at the

Secure Delivery
destination host, enforcing confidentiality
Network Service Models
Example: Internet’s Service Model

Guaranteed Delivery Internet’s service model is

Best Effort,
aka., no service at all!
Delay-bounded
Guaranteed Delivery

How to manage output queues in order

In-Order Delivery to implement Best Effort?

Minimal-Rate Delivery

Secure Delivery
Network Service Models
Example: Internet’s Service Model

Guaranteed Delivery Internet’s service model is

Is Best Effort a good network service model? Best Effort,
Does it work well? aka., no service at all!
Delay-bounded
Does it work the same way for everyone?
Guaranteed Delivery
We are about to discuss performance (or quality) of a
How
a fewtoperformance
manage output queues in order
In-Order Deliveryservice model. But first, let’s define to implement Best Effort?
indicators in networking.

Minimal-Rate Delivery

Secure Delivery
NETWORK TRAFFIC
CHARACTERISTICS
Characteristics of Network Traffic
Throughput Delay Jitter Loss
Characteristics of Network Traffic
Throughput Delay Jitter Loss
Definition: Link’s capacity, commonly (and wrongly) referred to as bandwidth, is the maximum data rate
achievable in a given link. Simply put, it is the link’s speed. Capacity is measured in bits per second (bps).

Example of capacity model is the Shannon-Hartley theorem:

𝑆
𝑅 = 𝐵∙ 𝑙𝑜𝑔2 1+
𝑁
R: Capacity [bps]
B: Bandwidth [Hz] What is the difference between
S: Transmission power [W] bandwidth and capacity?
N: Noise and Interference [W]
Characteristics of Network Traffic
Throughput Delay Jitter Loss
Definition: End-to-end throughput, is the speed, in bits/s, at which a client host receives data from the
server. It is bounded by the link with the smallest capacity in the communication path, which is called a
bottleneck link.

𝑇 = 𝑚𝑖𝑛 𝑅𝑠 ,𝑅𝑐

𝑇 = 𝑚𝑖𝑛 𝑅𝑖
𝑖=1,...,𝑁
Characteristics of Network Traffic
Throughput Delay Jitter Loss
Bottlenecks can also be formed when multiple data flows simultaneously share the link.

What is the throughput of each server-

client communication in this case?

Answer:
- If R is sufficiently large, R >> Rc and R >> Rs,
then:
T=min(Rc, Rs).

- Otherwise, considering that R is equally split

between the 10 data flows, then:
T=min(Rc, Rs, R/10)
Characteristics of Network Traffic
Throughput Delay Jitter Loss
Definition: The nodal delay is the time for a packet to be transmitted from one network node to the next
one. It the sum of the delay component at every communication stage, i.e.,

Processing Prop Transmission

Delay dNodal
i = dProc
i + dQueue
i,i+1 + dTrans
i,i+1 + di,i+1 Delay

(constant)
dProc L
i dTrans
i,j =
Ri,j
L: Packet size [bits/packet]

Queuing Propagation
Delay Delay

dQueue
i,j (later) Prop
di,j =
disti,j
c
c: Speed of light 3.0∙108m/s
Characteristics of Network Traffic
Throughput Delay Jitter Loss
Special case: The queuing delay is sensitive to the experienced traffic intensity ρi,j going through link (i,j).

dQueue
i,j
L∙a
ρi,j =
Ri,j

L: Packet size [bits/packet]

a: Average rate [packets/s]
Ri,j: Link (i,j) transmission rate [bps]

Traffic engineering design rule

0 ≤ ρi,j ≤ 1
Characteristics of Network Traffic
Throughput Delay Jitter Loss
Definition: The end-to-end delay (or one-way delay) is the time, in seconds (s), for a packet to get from the
source to a destination host. It is given by the sum of the nodal delay at all transmitting nodes in the data
flow Fs→c from server s to client c:

Queue Prop
dE2E = dNodal
i = dProc
i + d i,i+1 + d Trans + d
i,i+1 i,i+1
i∈Fs→c i∈Fs→c

The time it takes to get from a source to the destination and back is called the round-trip delay, commonly
referred to as RTT (round-trip time).

RTT = dNodal
i + dNodal
j
i∈Fs→c j∈Fc→s
Characteristics of Network Traffic
Throughput Delay Jitter Loss

Consider the network of the figure.

We observed that the amount of traffic sent by the
servers doubled and it caused a significant increase in the
end-to-end delay for the packets transiting from servers
to clients.

– What delay component (processing, queuing,

transmission, or propagation) is most likely to be the
cause of the end-to-end delay increase and why?
– Can you think of a way to reduce the end-to-end delay
in this case?
Characteristics of Network Traffic
Throughput Delay Jitter Loss
Definition: Instantaneous jitter is the variability of one-way delay between 2 consecutive packets in a data
flow (or stream of packets), given in seconds (s).

x3 x2 x1
Characteristics of Network Traffic
Throughput Delay Jitter Loss
Example:
x3 x2 x1

Data flow of N=3 packets.

E2E delay for packet i=1 is dE2E (1) = 20ms.

For packet i=2, dE2E (2) = 30ms. Jitter(2) = dE2E (2) − dE2E (1) = 30 − 20 = + 10ms (Dispersion)

For packet i=3, dE2E (3) = 10ms. Jitter(3) = dE2E (3) − dE2E (2) = 10 − 30 = − 20ms (Clumping)
Characteristics of Network Traffic
Throughput Delay Jitter Loss
Definition: Average jitter is the variance of one-way delay in a stream of packets in seconds (s).

Let X be a random variable capturing the inter-arrival time between two consecutive packets of the
same flow. Consider that its expected value Ε[X] and its variance Var[X] are known.
x3 x2 x1

In practice, if X={x1,..., xN} is the sample space containing the inter-arrival time of N packets, Jitter is
the sample variance S2X, i.e.,
S2X = 1 ∑ N
i=1 x i − x 2, where x = 1 ∑ N
i=1 x i
N N
Characteristics of Network Traffic
Throughput Delay Jitter Loss
Definition: The percentage of packets that are dropped due to full queues is called packet loss.

dQueue
i,j

The expected number of packets in the queue is

given by Little’s Law:
E[N] = a∙dQueue
i,j [# of packets in the queue]
Packet drop condition:
L∙a
a∙dQueue
i,j > Nmax ρi,jL∙a
=
Ri,j
Traffic types
Batch Application

File size: 103.92MB = 103.92x10⁶Bytes

Datagram size: 1.5x10³ Bytes
Datagram payload: 1.46x10³ Bytes

Nb. of datagrams: 71,179 datagrams

What characteristics are important?

• Throughput
• Delay
• Jitter
• Loss
Traffic types
Interactive Application

Simple Instruction-Response
Application: SSH

What characteristics are important?

• Throughput
• Delay
• Jitter
• Loss
Traffic types
Voice Application

Steps:
• Analog-Digital converter Codec
• Generate digital voice data
• Encapsulate data with IP/UDP/RTP headers

A few statistics:
• 1 s of voice data = 50 datagrams
• 24 - 80 kbps of required throughput
• Analog sound is digitized every 20ms

What characteristics are important?

• Throughput
How does this analysis change for voice
• Delay
and video applications?
• Jitter
• Loss
QUALITY OF
SERVICE
Fairness
• Fuzzy concept in networking. It may have multiple definitions, e.g.,
• All clients’ traffics are processed under the conditions (equal rules)
• All clients’ traffics are promised the same characteristics (equal experience)
• All clients’ traffics are preferentially treated (for example, according to
economic or geographical attributes)
• As network engineers, how to provide fair service to users?
 Fairness
Example: Shortest Path Routing

Applying the same rule, might provide

different performance levels to different C1 6
users.. 5
7

C2 8 9 4 C4

1 3

C3 2
S
Quality of Service
Definition: Quality of Service (QoS) is the description of performance
requirements for different traffic types. It also refers to the set of
• Architectures,
• Mechanisms, and
• Protocols
that are used to meet these requirements.

“Treating users differently to provide fairness

by meeting their requirements”
Quality of Service
Performance vs. Satisfaction
Quality of Service
Example of requirements – Voice traffic

Requirements:
• E2E delay < 150ms
• Jitter < 30ms
• Loss < 1%
Wrapping up

What did you learn today?

• Detailed review on L2 networking
and service models
• Introduction to the 4 traffic
characteristics important for QoS
• Definition of QoS