EP2200

Queueing theory and

teletraffic systems
Viktoria Fodor
Laboratory of Communication Networks
School of Electrical Engineering

Lecture 1
“If you want to model networks
Or a complex data flow
A queue's the key to help you see
All the things you need to know.”

(Leonard Kleinrock, Ode to a Queue

from IETF RFC 1121)

1
 What is queuing theory?
What are teletraffic systems?
Queuing theory
• Mathematical tool to describe resource sharing systems,
e.g., telecommunication networks, computer systems
– Requests arrive dynamically
– Request may form a queue to wait for service
• Applied probability theory

Teletraffic systems
• Systems with telecommunication traffic (data networks,
telephone networks)
• Are designed and evaluated using queuing theory

Why do we need a whole theory for that?

EP2200 - Lecture 1 2
Why do we need teletraffic theory?

• Waiting delays at output buffers

in network routers

• How will the delay change if the

number of packets arriving
within a second doubles?

EP2200 - Lecture 1 3
Why do we need teletraffic theory?

• Throughput (useful
transmissions) in a wireless
network with random access

• If transmissions may collide,

how would the throughput
change if the number of packets
to be sent doubles? What is the
effect of packet collisions?

EP2200 - Lecture 1 4
 Why do we need teletraffic theory?
Call blocking probability TCP throughput vs. packet loss
in a telephone network vs. load
0.8
0.4
Blocking probability

0.7

0.6

Throughput
0.3
RTT=t
5 lines 10 lines 0.5

0.2 0.4 RTT=10t

0.3

0.1 0.2

0.1

0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8 1

Load Experienced packet loss

• Teletraffic systems are non-linear, and mathematical tools are needed to predict
their performance

EP2200 - Lecture 1 5
 Course objectives
• Basic theory
– understand the theoretical background of queuing systems, apply
the theory for systems not considered in class

• Applications
– find appropriate queuing models of simple problems, derive
performance metrics

• Basis for modeling more complex problems

– advanced courses on performance evaluation
– master thesis project
– industry (telecommunication engineer)

• Prerequisites
– mathematics, statistics, probability theory, stochastic systems
– communication networks, computer systems

EP2200 - Lecture 1 6
 Course organization
• Course responsible
– Viktoria Fodor <[email protected]>
• Lectures
– Viktoria Fodor
• Recitations
– Ioannis (John) Glarapoulos
– Liping Wang

• Course web page

– KTH Social EP2200 (https://www.kth.se/social/course/EP2200/)
– Home assignments, project, messages, updated schedule and course
information
• Your responsibility to stay up to date!
– Useful resources: Erlang and Engset calculators, Java Aplets
– Useful links: on-line books
– Links to probability theory basics

EP2200 - Lecture 1 7
 Course material
• Course binder
– Lecture notes by Jorma Virtamo, HUT, and Philippe Nain, INRIA
• Used with their permission
– Excerpts from L. Kleinrock, Queueing Systems
– Problem set with outlines of solutions
– Old exam problems with outlines of solutions
– Erlang tables (get more from course web, if needed)
– Formula sheet, Laplace tranforms
For sale at STEX, Q2 building. Costs 100 SEK.

• No text book needed!

– If you would like a book, then you can get one on your own
• Ng Chee Hock, Queueing Modeling Fundamentals, Wiley, 1998. (simple)
• L. Kleinrock, Queueing Systems, Volume 1: Theory, Wiley, 1975 (well known, engineers)
• D. Gross, C. M. Harris, Fundamentals of Queueing Theory, Wiley, 1998 (difficult)
– Beware, the notations might differ

EP2200 - Lecture 1 8
 Course organization
• 12 lectures – cover the theoretical part
• 12 recitations – applications of queuing models
• Home assignments and project (1.5 ECTS, compulsory, pass/fail)
• Home assignment
– problems and (later) solutions on the web
– individual submission, only handwritten version
– you need 75% satisfactory solution to pass this moment
– submission on Nov. 20, submit at the STEX office
• Small project
– computer exercise
– details later
– submission deadline: Jan 4
– +5 points for outstanding projects

EP2200 - Lecture 1 9
 Exam
• There is a written exam to pass the course, 5 hours
– Consists of five problems of 10 points each
• Passing grade usually 20  3 points
– Allowed aid is the Beta mathematical handbook (or similar) and simple
calculator. Probability theory and queuing theory books are not allowed!
– The sheet of queuing theory formulas will be provided, also Erlang tables and
Laplace transforms, if needed (same as in the course binder and on the web)

• Possibility to complementary oral exam if you miss E by 2-3 points (Fx)

– Complement to E

• Registration is mandatory for all the exams

– At least two weeks prior to the exam

• Students from previous years: contact STEX ([email protected]) if you are not
sure what to do

EP2200 - Lecture 1 10
 Lecture 1
Queuing systems - introduction

• Teletraffic examples and the performance triangle

• The queuing model
– System parameters
– Performance measures
• Stochastic processes recall

EP2200 - Lecture 1 11
 Example
• Packet transmission at a large IP router
Input ports Output ports

Switching
engine

Routing
processor

• We simplify modeling
– typically the switching engine is very fast
– the transmission at the output buffers limits the packet forwarding
performance
– we do not model the switching engine, only the output buffers

EP2200 - Lecture 1 12
 Example
• Packet transmission at the output link of a large IP router –
packets arrive randomly and wait for free output link

• Performance:

• Depends on:

EP2200 - Lecture 1 13
 Example
• Packet transmission at the output link of a large IP router -
packets wait for free output link
• Performance:

• Depends on:

EP2200 - Lecture 1 14
 Example
• Voice calls in a GSM cell – calls arrive randomly and occupy a
“channel”. Call blocked if all channels busy.
• Performance

• Depends on:

EP2200 - Lecture 1 15
 Performance of queuing systems
• The triangular relationship in queuing

Service demand
stochastic

Server capacity Performance

stochastic
• Works in 3 directions
– Given service demand and server capacity  achievable performance
– Given server capacity and required performance  acceptable demand
– Given demand and required performance  required server capacity
EP2200 - Lecture 1 16
 Lecture 1
Queuing systems - introduction

• Teletraffic examples and the performance triangle

• The queuing model
– System parameters
– Performance measures
• Stochastic processes recall

EP2200 - Lecture 1 17
 Block diagram of a queuing system
• Queuing system: abstract model of a resource sharing system
– buffer and server(s)
• Customers arrive, wait, get served and leave the queuing system
– customers can get blocked, service can be interrupted

Interrupted
service
Arrival Completed
service
Blocking Under service
Return to Waiting
system

Exit system Buffer Server

EP2200 - Lecture 1 18
 Description of queuing
systems
• System parameters
– Number of servers (customers served in parallel)
– Buffer capacity
• Infinite: enough waiting room for all customers
• Finite: customers might be blocked
– Order of service (FIFO, random, priority)
• Service demand (stochastic)
– Arrival process: How do the customers arrive to the system –
given by a stochastic process
– Service process: How long service time does a customer
demand – given by a probability distribution
Customer:
• IP packet
• Phone call

EP2200 - Lecture 1 19
 Examples in details
• Packet transmission at the output link of a large IP router

• Number of servers: 1
• Buffer capacity: max. number of IP packets
• Order of service: FIFO
• Arrivals: IP packet multiplexed at the output buffer
• Services: transmission of one IP packet
(service time = transmission time = packet length / link transmission rate)

EP2200 - Lecture 1 20
 Examples in details
• Voice calls in a GSM cell
– channels for parallel calls, each call occupies a channel
– if all channels are busy the call is blocked

• Number of servers: number of parallel channels

• Buffer capacity: no buffer
• Order of service: does not apply
• Arrivals: call attempts in the GSM cell
• Service: the phone call (service time = length of the phone call)

EP2200 - Lecture 1 21
 Group work
• Service at a bank, with “queue numbers” and several clerks
• Draw the block diagram of the queuing systems

• Arrivals:
• Service:
• Number of servers:
• Buffer capacity:
• Order of service:

EP2200 - Lecture 1 22
 Performance measures
N
Nq Ns
• Number of customers in the system (N)
– Number of customers in the queue (Nq)
– Number of customers in the server (Ns)
• System time (T) W x
– Waiting time of a customer (W)
– Service time of a customer (x) T
• Probability of blocking (blocked customers / all arrivals)
• Utilization of the server (time server occupied / all considered time)

• Transient measures
– how will the system state change in the near future?
• Stationary measures
– how does the system behave on the long run?
– average measures
– often considered in this course
EP2200 - Lecture 1 23
 Lecture 1
Queuing systems - introduction

• Teletraffic examples and the performance triangle

• The queuing model
– System parameters
– Performance measures
• Stochastic processes recall

EP2200 - Lecture 1 24
 Stochastic process
• Stochastic process
–A system that evolves – changes its state - in time in a
random way
–Family of random variables
–Variables indexed by a time parameter
• Continuous time: X(t), a random variable for each value of t
• Discrete time: X(n), a random variable for each step n=0,1,…
–State space: the set of possible values of r.v. X(t) (or X(n))
• Continuous or discrete state

X(t) X(n)

time packet
• Number of packets waiting: • Waiting time of consecutive packets:
– Discrete space – Discrete time
– Continuous time – Continuous space

EP2200 Queuing theory and teletraffic 25

systems
 Stochastic process - statistics
• We are interested in quantities, like:
– time dependent (transient) state probabilities (statistics over many
realizations, an ensemble of realizations, ensemble average ):

f x (t )  P( X (t )  x), Fx (t )  P( X (t )  x)
– nth order statistics – joint distribution over n samples

Fx1 ,, xn (t1 ,, tn )  P( X (t1 )  x1 ,, X (tn )  xn )

x
– limiting (or stationary) state probabilities (if exist) :

f x  lim P( X (t )  x), Fx  lim P{ X (t )  x}

t  t 

ensemble average

EP2200 Queuing theory and teletraffic

26
systems
 Stochastic process - terminology
• The stochastic process is:
– stationary, if all nth order statistics are unchanged by a shift in time:

Fx (t   )  Fx (t ), t

Fx1 ,, xn (t1   ,, tn   )  Fx1 ,, xn (t1 ,, tn ), n, t1 ,, tn
– ergodic, if the ensemble average is equal
to the time average of a single realization
– consequence: if a process ergodic, then the
statistics of the process can be determined
from a single (infinitely long) realization and
vice versa
t

time average ensemble average

EP2200 Queuing theory and teletraffic 27

systems
 Stochastic process
• Example on stationary versus ergodic
• Consider a source, that generates the following sequences
with the same probability:
– ABABABAB…
– BABABABA…
– EEEEEEEE…
• Is this source stationary?
• Is this source ergodic?
t

time average ensemble average

EP2200 Queuing theory and teletraffic 28

systems
 Summary
Today:
• Queuing systems - definition and parameters
• Stochastic processes

Next lecture:
• Poisson processes and Markov-chains, the theoretical background to analyze
queuing systems

Recitation:
• Probability theory and transforms
• Prepare for the recitation: read Virtamo 1-3 in the course binder or
download from the course web
– Definition of probability of events
– Conditional probability, law of total probability, Bayes formula, independent
events
– Random variables, distribution functions (discrete and continuous)
– Z and Laplace transforms

EP2200 - Lecture 1 29