queuing. problem solving for performance evaluation analysis: use mathematical model to solve...
Post on 21-Dec-2015
221 views
TRANSCRIPT
Problem Solving for performance evaluation Analysis: use mathematical model to
solve usually need assumption to make it easy to solve Can have full domain of solution, I.e. optimal
solution Simulation: more realistic assumption
heuristic solution Analysis used to be the main research
area, now simulation is a popular method, but…
Queuing examples Customer service
Bank, post office , number of clerks vs. waiting time , service time
Computer servers and jobs (requests) Traffic flow on freeway or computer
networks.
Why probability ? Because everything is unpredictable. We
use ‘mean’ , ‘variance’ to be the metrics.
Sampling of Problems
Resource Scheduling Demands:
Unpredictable arrivals Unpredictable service demands Everybody wants immediate
service Conflict resolution
Waiting Loss sharing
Demand Resource
Single Resource with waiting (queuing)
Assume: Arrival Each service take Service
departures
serverqueue
Arrivals
)sec(customers λTT1rate arrival seconds every
seconds S
)sec(1 customersS
ccapacity
Single Resource with waiting (queuing)
Def: IF constant inter-arrival times AND
constant service times.
IF random inter-arrival times AND random service times.
. time response Average system) in (TimeR
SRc Rc
unstablec
Sampling of Problems
Every disk access is equally likely to be to any cylinder on the diskWhat is the average distance (in cylinders) of a seek ?
Expected execution time variance?
t1
branch
t2
t3
start
noexit
0.25
0.75
yes
Sampling of Problems
fi = fraction time link i is down at a random time , what is the prob. That there is
Cherry GOES DOWN TWICE A WEEK. What’s the prob. You lost your term report ?
S
f1
f2
f3
f4
f5f6
f7f8
f9
f10D
Probability Random experiment.
Real world experiment.outcome not predictable.
Associated with a random experiment. Set of all possible outcomes.
Mutually exclusive and exhaustive. Events
Event occurs or does not occur for each outcome. Defines a subset of outcomes.
occurs event then events if EOiOiE |
Probability
Independent trials If outcome of any trial does not affect
the outcome of any other trial If trials are performed under identical
conditions.
Probability
Statistical regularityn independent trials.
Def: likelihood differing
N as 0 to goes sexperiment of sets twofor than
moreby differing of likelihood
occurring.event A ofy probabilit
large For
NAn
AiAn
n
)(
)(~)(
(“ 點”機率不能用此法 )
Probability Sample space set of all possible outcomes
of a random experiment Not required, but often deal with equally probable outcomes.
Suppose event A occurs iffOi1 or Oi2 or … Oij occurs
NOi
jinOjn
nOinNiOi
1)(
,,)()(,...,2,1|
outcomes. possible all ofset Let
Probability
i.e.
jjOijA ,...,2,1|
J
j NJOij
J
j nOijn
n
J
jOijn
nAnAP
J
jOijnAn
1)(
1
)(1)(
)()(
1)()(
Probability Counting Things – Combinations
Product ruleLet A1 be a set containing n1 objects, A2 be a set containing n2 objects.The number of ordered pairs
(a1, a2) s.t is n1*n2
Generalizes tor sets A1,A2,…Ar of cordinalistics n1,n2,...,nrNumber of distinct r-tuples with i-th componemt an object from Ai is n1*n2*n3*…*nr
2211AaAa and,
Probability Some fundamental cases
Ways of selecting r out of n objectsTwo dimensions:
Ordered or unordered samples. (sequence v.s. a set)
With or without replacement. (whether can repeat an object)
Ordered sampling with replacement. The number of ways of selecting r objects in
succession from a set of n objects, where each object is returned prior to the next selection is nr
Probability Ordered sampling without replacement
The number of ways of selecting r objects in succession from a set of n objects (r ≤ n) where each object selected is removed from the set prior to the next selection is, n(n-1)…(n-r+1)= n!/(n-r)!
Sampling without replacement and without regard for order
The number of ways of selection r objects from a set of n objects without replacement & without regard for order is :
nrr
nrrn
n
, !)!(
!
Probability Sampling with replacement and without
regard for order r out of n
Ans:
r
rn 1000|0|00||00000…|000
1 2 3 4 5 …How many “0”s rHow many “1”s n-1
n
rrnyrx
r
nnyx0
)( : Theorem Binomial
r
n
r
n
r
n
r
n
r
n
nn
r
n
r
n
1
1
2
0)1(
0
0
Probability Probability Axioms
Let S be a sample space of a random experiment Axioms
(A1) for any event A,
(A2) P(S) = 1
(A3)
It follows easily that (n finite)
if Ai are mutually exclusive
0)( A
)()()( BPAPBAP exclusivemutually are B & A i.e. , whenever BA
n
iin APAAAP
121 )()...(
Probability
(A3’) For any countable sequence of events A1,A2,...,An,… , that are mutuall
y exclusive. (i.e. )
kjAA kj whenever
11)()(
nnnn
APAP
Some implications of the Axioms
1)()()()(
)(1)(
SPApApAApSAAAA
SAasAApAp
, :pf
w.r.t of complement where
0)( p
)()()()()()()(
)(:)()()()(
BApBpApBApBABpApBAp
BABABApfBApBpApBAp
Principles of inclusion & exclusion
If A1,A2,…,An are any events then
)...(1
)....()()(
)...()(
211
111
211
n-
nkjkji
njiji
n
ii
nn
ii
AAAP)(-
AAAPAAPAP
AAAPAP
Proof by induction on the number of events n (recitation)Related result in combinatoric S
Conditional Probability Probability that event A has occurred given
that B has occurred , notation: P(A|B) Def:
BSif
BSifBPSP
BSPi
ii
i
0)()(
)|(
Makes intuitive sense in terms of relative frequencies nBnn
iSn
)(
)(
)(
)(
)(
)(
)(
)(
)|()|()|(
BP
BAP
BP
SP
BP
SP
BSPBSPBAP
BASi
BAS
i
BAS BASji
i
i
i j
Generalization:
)|()|()(
)(
BACPABPAP
CBAP
Independent Events Def: Events A & B are said to be indepe
ndent iff In words, given that B occurred, given
no information about the probability that A occurred.
Note: If P(A|B)=P(A) , that
i.e. symmetric
)()|( APBAP
)()(
)|()(
)(
)()|( BP
AP
BAPBP
AP
BAPABP
Independent Events
Alternative Def:follows from first def: but if P(A|B)=P(A)
then P(A)P(B) = P(AB)
)()()( BPAPBAP
)(
)()|(
BP
BAPBAP
Independent Events
Ex:Pi=prob. of introducing an error, assume that n msg arrives in errorP(msg arrives with error)=?Let Ei=event that link I introduces an errorP(msg arrives with error)
assume are independent. are indep. ??=> P(msg arrives with error)
LA NYP1 P2
Kansas City
)(1
)(
21
21
EEP
EEP
21 E&E 21 E&E
))(1))((1(1
)(1
21
21
EPEP
EEP
Generalized
N
iiEP
1
))(1(1
Some random properties
If
If A & B are independent And B&C are independent
A&C are independent
0)(0)(
BPAP
BA
or then
t,independen are B & A and
Ex: B={ 1, 2, 3, 4 } A={ 1, 3, 5 } C={ 2, 4, 6}
Independence of a set of events
Def: A list of n events A1,A2,…,An is said to be mutually indep. iff for each set of k (2 ≤ k ≤ n) distinct indices i1,i2,…,Ik
pairwise indep. mutually indep.)(...)()()...( 2121 ikiiikii APPAPAPAAAP
Note: duality mutually exclusive :
mutually indep :
k
iik APAAAP
121 )()...(
k
iik APAAAP
121 )()...(
Ex: N blocks in a file. n accesses, uniformly distributed. What is the prob. That the first block
is accessed at least once ?(A) P[1st block accessed] =(B) #ways to make the n accesses = Nn
#ways with 0 accesses to 1st block = (N-1)n
#ways with >0 accesses to 1st block = Nn- (N-1)n
∴
n
n)
11(1
nnn
nn
)N1
(11)N
1N(1
N1)(NN
] P[
Let Ei = event that the first block is accessed on i-th accessP[1st block accessed at least once]
nn
i
n
ii
n
ii
n
ii
n
ii
NNEP
EP
EPEP
)1
1(1)1
1(1)(1
][1
][1][
11
1
11
Bayes Rule In general:
Let B1, … , Bn be mutually exclusive and exhaustive events.
Let A be any event.
kkk
jj
jj
BPBAP
BPBAP
AP
ABPABP
)()|(
)()|(
)(
)()|(
B1B2
B4
B3S A
jiBiB
SB
j
n
jj
,1
{Bi}exhaustive
Mutually exclusive
n
kk
n
kk
n
kk
BAP
BAPBAP
SAPAP
1
11
)(
))(())((
)()(
Ex: Drawers G G G SSelect a drawer at random reach in and take a coin without looking.The coin is gold, what is the prob that drawer 1 was chosen ?
Find P(drawer1|gold)=?P(drawer1)=p(drawer2)=1/2P(gold|drawer1)=1P(gold|drawer2)=1/2
3
2
4/3
2/1
2/1*2/12/1*1
2/1*1
drawer2)drawer2)P(|P(golddrawer1)drawer1)P(|P(golddrawer1)drawer1)P(|P(gold
gold)|P(drawer1
Chap 2 Discrete Random Variable
Def: A random variable X on a sample space S is a function X: SR from S to the reals Numbers often are naturally associated
with sample space outcomes. e.g. face value on a die
But not always e.g what if the die had colors
Probability mass function
Def: Px(x)=P(X=x)is also called the density function for X
(Cumulative) Distribution function
xSX
SP
xSXSPxXPAxP
)(
)(
}))(|({][)(
txX
X
xP
tXPtF
)(
)()(
1/36
3/36
2 3 4
Ex: Bruin rot1/10000 of having to disease Medical test.Two possibilities P (positive) or N (negative)P[P|BR]=0.99P[P|NBR]=0.01
Henry gets a test result is positive.P[BR|P] =
01.0*)101(99.0*1099.0*10
]|[][]|[][]|[][
44
4
-
nBRPPnBRPBRPPBRPBRPPBRP
Particular Cases Bernoulli
k]P[yΔP
xxxxY
q-p]P[X
pΔ] P[X
nk
inn
Let
r.v Bernoulli a is each
Binomial
1 & 0 valueswith R.V. discrete ,X
21
10
1
Particular Cases Experiments with two outcomes, repeat some num
ber of this n, There are 2n possible sequences of length n for outcomes of the experiments.
Call the outcomes “success” and “failure” Letfor any single experiment
P(success)=p P(failure)=1-p=q
Consider a sequence of length n 以下看不懂
Particular Cases Consider
knkknkk
k
qpk
nqpSeqpP
P
.)(
trivals n in success k of prob.
Seq. distinctSequence of length n With k successes
All seq with k success
All have the same prob
Particular Cases Generalization
m
m
nm
nn
m
mi
nm
nn
mm
PPPnnn
n
PPP
, P, , PP, O, , OO
m
21
21
2121
121
2121
)!!!
!(
nm)n2,P(n1,
trials. n of sequence have them
iesprobabilit with Say
trial of outcomesdifferent are there Suppose
Geometric r.v Z
(An) Interpretation: Number of Bernoulli trials until a success
is obtained sample space is infinite All sequences of some number of 0’s follow
ed by a 1 or fff….f s PZ(i)=qi-1*p , I >= 1
Geometric r.v Z Distribution functio
n Fz(t)=P(Z<=t) 字被擋住了,看不到
t
z
-q
s]st t trialess in fir-p[no succ
is event]ment of th-p[complet
is eventprob of th(t)F
ti
1
1
1
event an is with Sequences
Geometric r.v Z
stnyo"" ofset space sample
trials. first in success
no were therethat given success a
before beyond trials ofnumber
r.v. lConditiona
0
0
k
kr.vΔLet w
pqq
pq
]kp[z
]kkp[z
]kp[z
]kzkkp[z]k|zkkp[zk]p[w
k-k
kk1
1
0
0
0
0000
0
0
:property Memoryless
Poisson Pmf
k
n
k
kknk
nt
nt
n
t
kkn
n
n
t
n
t
k
nPk(t)
,t)(
λt/n as n
t/n
1
1
!)!(
!1
0 in events k ofy probabilit the
trial.t independen an is t/n interval each
interval an inevent an ofy probabilit the Suppose
這邊字看不清楚
Probability Generating Function (Z transform)
Transformation: Will simplify many calculations later.
(non negative integer valued r.v.)
xzG
pkdz
zGd
G
zpzG
pkxpLet
x
k
z
kx
k
x
k
kkx
k
of ondistributi complete
:2Property
:1Property
)(
!)(
1)1(
)(
)(
0
0
Probability Generating Function (Z transform)
An example
0 0
0 0
0
0
)()(
)()(
)(
)()()(
n j
jnj
n
n
j
n
nnz
n
j
zjnypzjxp
zjnypjxp
zpzG
jnypjxpnzppn
sum the i.e. y xz Define
n,0,1, range withr.v t independen bey Let x,
Some transforms
Bernoulli r.v.x
nx
knk
x
pzpzG
qpk
nkxp
pzppzqzzG
pxp
pqxp
)1()(
)(
1)(
)1(
1)0(
10
r.v.x binomial 這邊有字看不懂
Some transforms
Poisson
)1)(()1()1(
21
21
)1(
00
2121
21)()()(
&
!)()(
!)(
zzzxxz
z
k
zkk
k
k
k
eeezGzGzG
xxz
xx
eeezk
ezkXpzG
ek
kxp
21,parameter withr.v poison indep. two be Let
? know we do How
poisson is sr.v' poisson indep. two of sum
Linear recurrence relation
2
2 2 221
21
21
)(
11
102
n
nn
n n n
nn
nn
nn
nn
nn
nn
n
nnn
zfzGDefine
zfzfzf
zfzfzf
znth
f
fnfff
to 2n sum Now
by equation multiply
condition initial Let
接下頁
Linear recurrence relation
nnn
n n
nn
nn
n
nn
n
nn
BAz
z
B
z
A
zzzG
zfzffzzzG
zGzfzGz
zfzzfz
zfzzfzzffzG
21
212
0102
20
1 0
2
2
22
2
2
1110
111
1)(
]1)[(
)(])([
)(
Review
pq
p
xXpxF
xx)(xp
xxx
RSX
X
iix
1
][)(
,2,1
success of prob Bernoulli
fun. onsdistributi
prob
valueson takes
fun. mass prob
r.v Discrete
:
:
Review
pqp
vr
qpC
kpk
kk
knknk
1
.
successfirst until trials Bernoulli of #
Gexxxxxx
n) of(out success of prob
trials Bernoulli Binomial
:
:
字看不懂
這邊的也看不懂
Z-transform Fn n=0,1,2,3… Define Property 1:
Property 2:
0
)(n
nn zfzF
fun. mass prob a repress }{1)1( fnifF
0
)(
!
1
z
k
k
k dz
zFd
kf
Z-transform
i
i
n
ek
kXp
tX
zFzFz
Xz
nix
zFzFzFyxz
zFzFyx
ki
i
ii
n
ix
i
i
YXZ
YX
!)(
)()(
,,2,1
)()()(
)()(
1
)( parameter with r.v Poisson 2.
indep.mutually are , generalize
transform-z has r.v. the then
& transforms-z withr.v indep. are & If 1.
續下頁
Z-transform
k
j
jkj
jkk
j
j
k
j
zzzY
ej
k
k
jk
e
j
e
jkxpjxpkYP
eeezF
XXY
0
)(21
2
0
1
0
)1)(()1()1(
21
21
21
2121
!
1
)!(!
)2()1(][
)(
Let
下面好像還好,看不清楚
Z-transform kleinrock Appendix I
a
zF
a
f
z
zFf
zFzffdz
zdFznf
fzFz
f
azFfa
zFfn
n
n
kn
nn
n
n
nn
)(
1
)(
)()1(
)(
))((1
)(
)(
0
1
01
Z-transform
32
2
2
)1(
)1(2
)1(
)1(
1
101
1
1,2,1,01
0
1
100
01
z
zn
z
zn
z
zn
z
AA
z
zknU
znU
zkn
kn
n
n
n
n
k
kn
n
kkn
n
Last 4 are special cases of the following
Z-transform
)(
)()(
)(
!
1
!
1
1)1(
1)1()1)((
!
1
0
1
zD
zNzF
dz
zFd
nf
en
mz
nmnmnm
z
n
n
n
z
mn
Expansion Fraction Partial
theorem) valueate(Intermedi methodpower
inversion) of (method
,
p.335
Discrete random vectors Multivariate distributions
16
6,3
16
5,2
16
5,1
)(
),,,()(
),,,(
2211
21
pmf marginal
r.v. are where
x
x
x
xp
xXxXxXpxXp
xXXXX
X
nn
nn
1 2 3
1
2
X
Y
1/4 1/4 1/4
1/16 1/16 1/81/4
3/4
),( yxpXY
續下頁
Discrete random vectors What is the condition pmf )2/( yixp
z
j
jzypjxpzZp
yxDef
0
)()()(
:
nconvolutio (discrete)
YXZ old of terms in r.v. new Defining
spmf' marginaltheir ofproduct the is pmf
joint their ifft independen are and
Continuous r.v
x
XXx
X
X
dttfxFdx
xdFxf
xFx
xxXpxF
)()()(
)(
)(,
)()(
or
fun.density y Probabilit
.continuous is r.v. continuous afor
: Function onDistributi
FX(x)
x0 x0
Limit=0
Limit=1
FX(x)
xx0 a b
字看不懂nondecreasing
Application to simulation
How to generate a r.v. with desired FX(x) 先 generate a uniformly distributed r.v. y Then x0=FX
-1(y0) got it Nonnegative r.v
)()]([][
)(
][][
000
00
00
xFxFypxxp
xFy
yypxxp
XX
X
x
y
x0
y0
這邊有字看不到
Exponential distribution Service time at a resource (server) Distribution of time between arivals
00
0)()(
01)()(
x
xe
dx
xdFxF
xexFxXpx
x
,
,
,
function ondistributi
x
F(x)
Exponential distribution
)()(1
1)( ytf
tFf x
Xyt
get a ofcomponent a of life remaining
component new a of time life
tY
X
Fx(x)
xtRemaining life distribution
)(1
)(
)(1
)()(
)(
)(1
)()(
][
][
]|[][)(
tF
ytfx
tF
ytFdyd
dy
ydFyf
tF
tFytF
tXp
ytXtp
txytXpyYpyF
XX
Xy
y
X
XX
ty
t
t
t
Exponential distribution Memoryless property
yt
ty
t
ty
X
Xx
xX
ee
e
e
e
tf
tyfyf
exf
xxp
xxxxpxxxxxp
t
)()(
0
0000
)1(1)(1
)()(
)(
)(
)()|(
Poisson distribution of exponential distribution
Poisson distribution was for the xxxxxxx of “events” in an interval of length t
where
0)(
lim)(
)(1[
)(10[
)(1[
t
tOttO
tOp
tOtp
tOtp
Δt
0ts.t of functionany is
t] inevent
t] inevent
t] inevent
length of lsubinterva eachfor
Poisson distribution of exponential distribution
(a) Exponential interarrival
(b) Poisson arrival process memoryless property exponential interarrival times
What is the “memoryless property” important ?Real answer must wait but intuitively.
)(
2
))((
)2
)(1(1
1),([
tO
t
tt
tt
etttp
in arrival
Arrival process departures
Expectation Def of expectation
X: a random variable
Moments
r.v continous
r.v discrete valueexpectedor mean
dxxxf
xpxxE i
ii
)(
)(][
dxxfx
xpxEYE
xY
ii
)()(
)()]([][
)(
Let
Expectation
2meanvariance
sequared varianceoft coefficien
deviation Standard
Variance
or
moment Second
22
222
]])[[(
)(][
xExE
pxdxxfxxEi
ix
xi
Expectation Expectation is a linear operation
Variance Xi are mutually indep.
][][
,,,
][][
1
21
i
n
ii
n
XEYE
XY
xxxn
bxaEbaxE
variablesrandom of Sum
N
i
N
ixii
x
jijjii
N
iii
N
iii
N
iii
i
i
i
xxE
xxxxxxE
xxExxEyyE
XY
1 1
22
0
1
2
2
1
2
1
2
])[(
]))((2])([
]))([(])[(])[(
indep. is if
Expectation Expectation of functions of more than one random variable nxxx ,,, 21
disance of valueexpected find towant
distance
disk afor time seek expected the
or
|-X|X
Example
xxpxx
dxdxxxfxxYE
xxY
nxnn
x
nnn
n
12
11
111
1
:
),,(),,(
,),,(),,(][
),,(
1
0 aNextrequest
currentrequest
x1x2
Conditional Distribution and Conditional Expectation
Conditional Probabilities
xXYX
xY
XYX
x
XY
xypxpyxpyp
xypxpyxp
xp
yxp
xXp
xXyYpxXyYpxyp
yxpYX
Bp
BApBAp
)|()(),()(
)|()(),(
)(
),(
][
][]|[)|(
),(&
)(
)()|(
|
|
|
Yfor pmf marginal the And
then Also
pmf with (discrete) r.v.dependent be Let
Marginal pmf
Conditional Distribution and Conditional Expectation
Continuous r.v.
)|()(),(
)(
),(
][
][
]|[)|(
)(
),()|(
|
|
|
xyfxfyxf
dxxf
dxdyyxf
dxxXxp
dxxXxdyyYyp
dxxXxdyyYypdyxyf
xf
yxfxyf
XYX
X
XY
XXY
yprexxxxxxl As
xxxxxxxxxIn 看不懂的字
看不懂的字
續下頁
Conditional Distribution and Conditional Expectation
)(
),(),()|(
)|()(),()(
||
|
xf
dttxfdttxfxyF
dxxyfxfdxyxfyf
X
y
y
XYXY
XYXY
Function onDistributi lConditiona
這邊的三個 example 字我看不太懂,先跳過了…
Conditional Expectation
]][[
)(][
)()|(
)|()(
)(][
)(
),()|()|(
]|[]|[
|
|
|
)(
|
|
YEE
dxxfYE
dxxfdyxyyf
dydxxyfxfy
dyyyfYE
xf
dyyxyfdyxyyfxypy
xYExXYE
XYX
XXY
XXY
yf
XYX
Y
XyiXYi
Y
i
nExpectatio total of Thm
:Note
or
or
- -
Conditional Expectation
pq
pqXE
pqXE
UpUXEUpUXEXE
U
UE[V]UXE
VUX
pkqXE
X
k
k
1
1][
*1])[1(
]1[]1|[]0[]0|[][
11
01]|[
][
:
1
1
r.v. distri. geometric :Ex
1st begin required
trials of #trial1st
on success
If U=1 then V=0If U=0 then V has same pmf as X
=1+E[X]
Example
timerepair the during building willthat
time) execution of terms (in backlog mean the Find (a)
hours 6b & hours 3a
between ddistribute uniformlly be will time
repair the and down goes system thethat Assume
parameter with lexponentia and
i.i.d are jobs the of time execution thethat Assume
,
arrivethat jobs ofnumber klet t, interval in i.e.
distr. Poisson a to according arrive jobs Suppose
0!
)()( k
k
etkKp
tk
time
1 n2 …… exeuction time
Repair time
Jobs arrivals
nd,ND)T T(T
T
NiTnNdDTE
TTTTTE
TTT
T
N
D
N
i
N
iiii
N
N
given for density need
nalunconditio
variablerandom of sum Random ,
job individual the of time execution
time execution total
period down in arriving jobs ofnumber
length period downLet
21
1
21
21
.1
]1,,,|[
][
,,,
求
1
],|[
),|(
00
22
0
11
0
211
21,|
21
21
21
21
n
dedede
dddeeenNdDTE
eee
nNdDP
nn
n
N
ii
nnNdDTTT
n
n
n
N
Laplace Transforms
)(*)(:0
sFtfNotation
f(t)]E[edtf(t)e F*(s)
sdtf(t)eΔF*(s)
tf(t) , -
-sT-st
st
density r.v.with the is T when ,
functiondensity the be will f(t) and
variablesrandom enonnegativ with deal will we
riablecomplex va is ,
as defined is f(t) of ransform Laplace the
function aFor
Some properties
)()()()(
)()(
)()1(
)(
)(
)(
)()(
)(0)(
)(*)()()(
)()(0)(
**
11*
*
*
*
*
*
0
*
1
sGsFtgtf
dssFt
tfds
sFdds
sdF
tft
ttfmoments
sFeatf
asaFaatf
sbGsaFtbgtaflinear
dtetfsFttf
nconvolutio
ss
n
nn
n
as
st
,
,
Transform Function
0
*
*
)1(21*
*
0
*
)()0(
)(
)0()0(')0()()(
)0()()(
)()(0)(
dttfFpropertyIntegral
s
sFf(τ(τ)
ffsfssFsdt
tfd
fssFdt
tdf
dtetfsFttf
t
-
nnnnn
n
st
,
Transform Function
Laplace transform pairs
1
2
0
0
)(
1)(
!
)(
1)(
)(
)(
1
1
nat
n
at
at
as
-as
astue
n
t
astute
as
AtuAe
s
eatu
s u(t)
ea)(t δ
(t) δ
Transform Function
Why we need transforms ?
Ex: Random sum of random variables
)()(
)(
)(][][)(
)(]|[
)()(
21
**
*
0
**
*
*
1
sXNsT
sXN
sXnNpeEsT
sXnNeE
sXXZNN
X
N
XT
n
nsT
nsT
i
N
ii
(s)T T,for L.T Find
L.T has , transform-z has
distri.)y identicall & (indept. i.i.d. are
,, r.v. discrete is when
*
z-transform
(continuous)
Inversion of Laplace transforms
1. Explicit inversion formula -- we won’t use
2. Inspection – using table3. Partial fraction expansion
p)-(1 prob ,1(
p) prob ,0(
p-1 probwith
p probwith
X
XZ
r.v. a is ZSuppose ..
][
X r.v a of ..
Transform Laplace
nExpectatio lConditiona
2
1
Y
Y
ge
eE
TL
Φ(X,Y)EE
(x,y)dxdyΦ(x,y)fE[Φ[Φ(X,Y
of X & Y function (X,Y) be abe r.v, ΦLet X & Y
SX
X|YY
XY
)()1()(
eE1eE0Yp
eEE
][)(
X of L.T. thebe (s)F
X of L.T. thebe (s)F
**
SX-1Y|X
SX-0Y|X
Y),X,(Xs-YZ|Y
*
2*X
1*X
21
2
2
1
1
21
2
1
sFpspF
Yp
eEsF
Let
XX
szZ
Random Sum of random variables
n*X
sX-X
sX-X
sX-X
sX-sX-sX-nN|Z
sZ-nN|Z
i
N
ii
(s)F
][eE][e]E[eE
]ee[eE][eE
Zfor L.T. the wantWe
times execution individual are X ,XZ
worktotal ZLet
s)(continuou X r.v. a as
ddistribute time execution an has job each (discrete) N r.v. a is
systema to submittedjobs of number theday a over Ex
n
n
2
2
1
1
n21
1
:
sometimesinvert can we
? it withdo wedo What
sFGsF
nNpsF
nNpeEeE
XNZ
n
nX
n
sZnNZ
sZZ
)()(
][)]([
][][][
**
0
*
0|
Z)λr(
rN|R
λrn
*Z
e(Z)G
en!
(λλrr]n|Rp[N
r) parameter (Poisson arrive that jobs of numberN Let
sF
repair of end the at workof backlog total ZLet
X(i.i.d) as dist. time execution an have job each
parameter process Poisson a are arrivals Job
R r.v. a is time Repair
fails systemcomputer Ex
1
?)(
:
))(1((
)(
)()()(
)(
*
0
))](1([
0
*|
*
))](1(
))(1(*|
*
*
*
sFF
drrfe
drrfsFsF
e
esF
X*R
RsFr
RrRZZ
sFr[-
sFrrRZ
X
X
X
Last-come-first-serve
nsb
sgsgsb
gggbs
sz-
*
*
sGe
eEeEeE
bBnNeE
bBnNE[e
time totalz Let
arrivals of numberN
time total for L.T(s)G
(s)Btime serviceof TL
n
n
)]([
][][][
],|[
],|
.
*
)(
1
21
))(()(
)(][
]|
***
0
))((
))(
))(1(
0
*
*
*
sGsBsG
dbbfeeE
e
ee
(s)][Gb]eB|np[NbBE[e
N on nalunconditio
BsGsbsz
sG-b(s-
sGbsb
n
n*sb-sz-
Review of Probability
N
A in outcomes #
outcomes possible of
number totalN where ,
Np(A)
outcomeslikely Equally
relative : regularity lstatistica
trials Indep.
outcome} possible a is o{o space sampleoutcomes,
experiment Random
Ao
ii
i
1
|
1-n
1-rn
order for regard withoutand treplacemen with
r
n
r)!-(nr!
n!
order for regard withoutand treplacemen without
r)!-(n
n! treplacemen without samplesordered
n treplacemen with samples(2)ordered
rule Product
nCombinatio
r
)5(
)4(
)3(
)1(