A bit on Queueing Theory:M/M/1, M/G/1, GI/G/1
Yoni Nazarathy*
EURANDOM, Eindhoven University of Technology,The Netherlands.
(As of Dec 1: Swinburne University of Technology, Melbourne)
Swinburne University Seminar, Melbourne, July 29, 2010.
*Supported by NWO-VIDI Grant 639.072.072 of Erjen Lefeber
Outline
• The term: queueing theory• The single server queue• M/M/1, M/G/1, GI/G/1• Mean waiting time formulas• Derivation of the M/M/1 result• A glimpse at my queueing research
The Term: Queueing Theory
Queues• Customers:
– Communication packets– Production lots– Customers at the ticket box
• Servers: – Routers– Production machines– Tellers
• Queueing theory:– Quantifies waiting/congestion phenomena– Abstract models of reality– Mostly stochastic– Outputs:
• Performance evaluation (formulas, numbers, graphs)• Design and control (decision: what to do)
Queueing Research• 1909: Erlang – telephone lines• Dedicated journal: Queueing Systems• Other key journals:
– Stochastic Models– Applied Probability Journals (JAP/Advances)– Annals of Applied Probability– OR, ANOR, ORL, EJOR…– About 5 other applied probability journals
• Books: Around 200 Teaching/Research• Active researchers: ~500• Researchers that “speak the language”: ~2000• Related terms: “Applied Probability”, “Stochastic Modeling”
Queueing Theory Applied in Practice
• Here and there…– Practice motivates many new queueing problems– BUT: Queueing results not so often applied– Accurate data sometimes hard to obtain– Models are often too simple for very complex realities
• Simulation can do much more…• …but say much less• Insight gained from queueing theory is important
The Single Server Queue
The Single Server QueueBuffer Server
0 1 2 3 4 5 6 …Number in
System:
A Single Server Queue:
( )Q t Number in system at time t
( )Q t
t
The Single Server QueueBuffer Server
0 1 2 3 4 5 6 …Number in
System:
A Single Server Queue:
{ , 1}nT n Arrivals times
{ , 1}ns n Service requirements1{ , 1}n n nt T T n Inter-Arrivals times 0 0T
The sequence ( , ), 1n nt s n Determines evolution of Q(t)
( )Q t Number in system at time t
( , ), 1n nt s n
( )Q t
nW
nW The waiting time of customer n
1 1 ,0n n n nW Max W t s
Performance Measures
Some important performance measures:
( )nP W x lim [ ]nnW E W
Little’s result: L W
lim ( )t
L E Q t
1[ ]nE t 1[ ]nE s
Assume the sequence is stochastic and stationary
Load
Stable when 1
We can quantify L (or W) under some further assumptions on
( , ), 1n nt s n
M/M/1, M/G/1, GI/G/1
Notation for Queues• A/B/N/K– A is the arrival process– B is the service times– N Is the number of servers– K is the buffer capacity (default is infinity)
M/M/1, M/G/1, GI/G/1
• M Poisson or exponential or memory-less• G General• GI Renewal process arrivals
( , ), 1n nt s nAssumptions on :
0
xtP X x e dt
Results for Mean Waiting Time
Mean Waiting Time1
/ /1 1M MW
21
/ /11
1 2s
M GcW
2 2 21
/ /1 21 2a s
GI Gc cW
2 2
2 2
( ) ( ),n na s
n n
Var t Var sc cE t E s
Derivation of the M/M/1 Result
A Markov Jump Process
0 1
2
Due to M/M (Exponential), at time t, Q(t) describes the state of the process
( )( ) ( ) ( ) 1 ( ) 1 , 0,1, 2,...
dP Q t jP Q t j P Q t j P Q t j j
dt
lim ( )j tP Q t j
, 0j j Stationary distribution
0j
j
L j
01 Utilization
The Stationary Distribution
0 1
2
(1 ) jj
1, 0,1,2,...j j j
0
1jj
1L
01
Solution:
Performance measures:
lim ( ( ) ) j
tP Q t j
My Research
During PhD
• Control and stability of Queueing Networks
• Queueing Output Processes
2 2
/ /1/ , =1
2 2
21Var ( )lim
1 (1)3
a sGI G K
x
a s K
c c KD t
tc c o K
During Post-doc
• BRAVO Effect(Seminar tomorrow at Melbourne University)
• Sojourn Time Tail Asymptotics• Methods of Control Theory Applied to Queues• Stability of Queueing Networks• Asymptotic scaling of stochastic systems• Optical Packet Switching Applications
In future…
• Research area: Model selection and statistics of queueing networks (from data)
• Engineering applications• More on previous subjects• Power supply networks
Thanks for Listening and See you Dec 1, 2010