Queuing Theory
Download
Report
Transcript Queuing Theory
Probability Review
Thinh Nguyen
Probability Theory Review
Sample space
Bayes’ Rule
Independence
Expectation
Distributions
Sample Space - Events
Sample Point
Sample Space S
The outcome of a random experiment
The set of all possible outcomes
Discrete and Continuous
Events
A set of outcomes, thus a subset of S
Certain, Impossible and Elementary
Set Operations
Union A B
Intersection A B
Complement AC
Properties
S
A B
AC
Commutation
A B B A
Associativity
A B C A B C
Distribution
A B C A B A C
De Morgan’s Rule
C
A B AC B C
A B
Axioms and Corollaries
Axioms
0 P A
PS 1
If A B
P A B P A P B
If A1, A2, … are pairwise
exclusive
P Ak P Ak
k 1 k 1
Corollaries
P AC 1 P A
P A 1
P 0
P A B
P A P B P A B
Conditional Probability
Conditional Probability of
event A given that event
B has occurred
P A B
P A | B
P B
If B1, B2,…,Bn a partition
of S, then
A B
(Law of Total Probability)
A B
AC
B1
B2
P A P A | B1 P B1 ...
P A | B j P B j
S
A
B3
Bayes’ Rule
If B1, …, Bn a partition of S then
P B j | A
P A B j
P A
P A | B j P B j
n
P A | B PB
k 1
posterior
k
k
likelihood prior
evidence
Event Independence
Events A and B are independent if
P A B P A P B
If two events have non-zero probability and are mutually
exclusive, then they cannot be independent
Random Variables
Random Variables
The Notion of a Random
Variable
The outcome is not always
a number
Assign a numerical value to
the outcome of the
experiment
Definition
A function X which assigns
a real number X(ζ) to each
outcome ζ in the sample
space of a random
experiment
S
ζ
X(ζ) = x
x
Sx
Cumulative Distribution Function
Defined as the probability
of the event {X≤x}
FX x P X x
Fx(x)
1
Properties
0 FX x 1
x
lim FX x 1
Fx(x)
lim FX x 0
¾
x
x
if a b then FX a FX a
P a X b FX b FX a
P X x 1 FX x
1
½
¼
0
1
2
3
x
Types of Random Variables
Continuous
Probability Density
Function
fX x
FX x
dFX x
dx
x
f X t dt
Discrete
Probability Mass
Function
PX xk P X xk
FX x PX xk u x xk
k
Probability Density Function
The pdf is computed from
fX(x)
dFX x
fX x
dx
Properties
P a X b f X x dx
b
fX(x)
a
FX x
x
f X t dt
1
f X t dt
For discrete r.v.
f X x PX xk x xk
k
dx
P x X x dx f X x dx
x
Expected Value and Variance
The expected value or mean of
X is
E X tf X t dt
E X xk PX xk
2
Var X 2 E X E X
k
The standard deviation of X is
Std X Var X
Properties
E c c
The variance of X is
Properties
E cX cE X
Var c 0
E X c E X c
Var cX c2Var X
Var X c Var X
Queuing Theory
Example
nSend
a file over the internet
packet
Modem
card
buffer
link (fixed rate)
place
C
Computation
(Queuing)
transmission
propagation
Delay Models
B
A
time
Queue Model
Practical Example
Multiserver queue
Multiple Single-server queues
Standard Deviation impact
Queueing Time
Queuing Theory
The theoretical study of waiting lines,
expressed in mathematical terms
input
server
queue
Delay= queue time +service time
output
The Problem
Given
One or more servers that render the service
A (possibly infinite) pool of customers
Some description of the arrival and service
processes.
Describe the dynamics of the system
Evaluate its Performance
If there is more than one queue for the server(s), there may also
be some policy regarding queue changes for the customers.
Common Assumptions
The queue is FCFS (FIFO).
We look at steady state : after the system has
started up and things have settled down.
State=a vector indicating the total # of customers in
each queue at a particular time instant
(all the information necessary to completely describe
the system)
Notation for queuing systems
A/B/c/d
A = the interarrival time distribution
B = the service time distribution
c = the number of servers
d = the queue size limit
nomitted
:Where A and B can be
D for Deterministic distribution
M for Markovian (exponential) distribution
G for General (arbitrary) distribution
if infinite
The M/M/1 System
Poisson
Process
Exponential
server
queue
output
Arrivals follow a Poisson process
nReadily
amenable for analysis
nReasonable
for a wide variety of situations
a(t) = # of arrivals in time interval [0,t]
= mean arrival rate
t = k ; k = 0,1,…. ; 0
Pr(exactly 1 arrival in [t,t+]) =
Pr(no arrivals in [t,t+]) = 1-
Pr(more than 1 arrival in [t,t+]) = 0
Pr(a(t) = n) = e- t ( t)n/n!
Model for Interarrivals and Service times
Customers arrive at times t0 < t1 < .... - Poisson distributed
The differences between consecutive arrivals are the
interarrival times : n = tn - t n-1
n in Poisson process with mean arrival rate
exponentially distributed,
, are
Pr(n t) = 1 - e- t
Service times are exponentially distributed, with mean
service rate
:
Pr(Sn s) = 1 - e-s
System Features
Service times are independent
service times are independent of the arrivals
Both inter-arrival and service times are memoryless
Pr(Tn > t0+t | Tn> t0) = Pr(Tn t)
future events depend only on the present state
This is a Markovian System
Exponential Distribution
given an arrival at time x
P (x (x t ))
P (( x ) t| x )
P ( x )
) (1 e
P ( (x t )) P ( x ) (1 e
1 P ( x )
P ( x )
(x t )
(x t )
x
x
t
e (1 e )
e
e
x
x
e
1 (1 e )
Same as probability starting at time = 0
x
)
Markov Models
• n+1
Buffer
Occupancy
departure
•n
• n-1
t
•n
arrival
t
Probability of being in state n
Pn (t t ) Pn (t )[(1 t )(1 t ) tt ]
Pn 1 (t )[(t )(1 t )]
Pn 1 (t )[(t )(1 t )]
as t 0, Taylor series
dPn (t )
Pn (t t ) Pn (t )
t
dt
Steady State Analysis
Substituting for Pn (t t )
( ) Pn Pn 1 Pn 1
Steady state
P0 P1
Markov Chains
0
1
...
n-1
n
n+1
Rate leaving n = Pn ( )
Rate arriving n = Pn 1 Pn 1
Steady State Pn ( ) Pn 1 Pn 1
State 0 P0 P1
Substituting Utilization
P1 P0 P0
P2 ( ) P1 P0
P2 P1 P1 P0 P1 ( 1) P0
Substituting P1
P2 P0 ( 1) P0
P0 P0 P0 P0
2
2
Pn P0
n
• Higher states have decreasing probability
• Higher utilization causes higher probability
of higher states
What about P0
P0
Pn 1 P0 P0
n 0
n 0
n 0
1
P0
1
P0 1
1
n
Pn (1 ) n
Queue determined by
n
E(n), Average Queue Size
q E (n) nPn n(1 ) (1 ) n
n
n 0
=
1-
n 0
n 0
n
Selecting Buffers
E(N)
1/3
1
3
9
.25
.5
.75
.9
For large utilization, buffers grow exponentially
Throughput
Throughput=utilization/service time = /Ts
For =.5 and Ts=1ms
Throughput is 500 packets/sec
Intuition on Little’s Law
If a typical customer spends T time units, on the overage, in
the system, then the number of customers left behind by
that typical customer is equal to
q Tq
Applying Little’s Law
M/M/1 Average Delay
E (n) E (T ) or w Tw or q Tq
E ( n)
1
1
E (T )
(1 ) (1 )
Ts
1
/
Ts so E (T )
(1 ) (1 ) (1 )
Probability of Overflow
n N 1
n N 1
P (n N ) pn (1 ) n N 1
Buffer with N Packets
1
p n 1 p 0 p0
n 0
n 0
1
n
1
(1 )
p0
and pn
N 1
N 1
1
1
N
N 1
N
n
(1 )
N
N +1
pN
(1 ) with 1
N 1
1
N
Example
Given
Arrival rate of 1000 packets/sec
Service rate of 1100 packets/sec
Find
1000
0.91
1100
Utilization
Probability of having 4 packets in the queue
P4 (1 ) .062
P1 .082, P2 .075, P3 .068, P5 .056
4
Example
E ( n)
9.99
1
With infinite buffers
P (n 12) 121 .28
With 12 fixed buffers cell loss probabilit y
( 1 )
P12
.04
121
1
Pn .11,.10,.09,.09,.08,.07,.07,.06,.05,.05,.04
12
Application to Statistcal Multiplexing
Consider one transmission
line with rate R.
Time-division Multiplexing
Divide the capacity of the
transmitter into N channels,
each with rate R/N.
R/N
R/N
R/N
1
T
Statistical Multiplexing
Buffering the packets
coming from N streams into
a single buffer and
transmitting them one at a
time.
R
T'
1
T
N N N
Network of M/M/1 Queues
1
1
3
4
2
3
2
1 1 2
Li
i
i i
2 1 2 3
3 1 3
1 2 3
L L1 L2 L3
T
1
i
J
i 1
i
i
M/G/1 Queue
The customers pat at rate C since each
customer pays C on the average and
customers go through the queue per unit time.
Assume that every customer in the queue pays at rate R when his or
her remaining service time is equal
R. time t, the customers pay at a rate
At atogiven
equal to the sum of the remaining service times
of all the customer in the queue. The queue
begin first
S 2come-first served, this sum is equal to
SQqueueing
Total cost paid by a customer: the
time of a customer who would
2
2
enter the queue at time t.
E
[
Q
]
E
[
S
]
Expected cost paid by each customer: C
S : Service Time
Q : Queuing Time
S
0
Q
2
E[Q] E[ S 2 ]
E[Q] C
2
E[S 2 ]
E[Q]
2(1 )
1
T E[Q]
S