Chap3f Secure PDF

Probability and Statistics with
Reliability, Queuing and Computer

Science Applications:
Second edition
by K.S. Trivedi
Publisher-John Wiley & Sons
Chapter 3: Continuous Random Variables

Dept. of Electrical & Computer Engineering
Duke University
Email: kst@ee.duke.edu
URL: www.ee.duke.edu/~kst
Copyright 2006 by K.S. Trivedi
Definitions
Distribution function:
If FX (x) is a continuous function of x, then X is a
continuous random variable.
FX (x) : grows only by jumps Discrete rv
F (x ) : both jumps and continuous growth Mixed rv
X
Note
We will also allow defective distributions. Defective

distributions, also known as improper distributions will be
covered later and are very useful in computer science
applications
These distributions satisfy F1, F2 and a modified version of
F3:
Unless otherwise specified, we will assume all distributions

to be non-defective
Definitions
(Contd.)
Equivalence:
CDF (Cumulative Distribution Function)
Probability Distribution Function (PDF)

but avoid this name as it can be confused with
pdf (prob. density function)
Distribution function
FX (x) or FX(t) or F(t)

probability density function (pdf)
X : continuous rv, then,
CDF and pdf can be derived from each other
pdf properties:
Definitions
(Continued)
Equivalence: pdf
probability density function
density function
density
dF
f(t) =
dt
F (t ) =
f ( x ) dx
= f ( x ) dx
0
, for a non-negative
random variable
Example 3.1
Random variable X : time (years) to complete a project
fX clearly satisfies property (f1).

To be a pdf, it must also satisfy (f2),
Prob. of completing project in less than 4 months,
Exponential Distribution
Arises commonly in reliability & queuing theory.

A non-negative continuous random variable.
It exhibits memoryless property.
Related to (discrete) Poisson distribution
Often used to model
Interarrival times between two IP packets (or voice
calls)
Service time distribution
Time to failure, time to repair etc.
The use of exponential distribution is an assumption

that needs to be validated based on experimental
data; if the data does not support the assumption,
other distributions may be used
For instance, Weibull distribution is often used to
model time to failure; Markov modulated Poisson
process is used to model arrival of IP packets
Mathematically (CDF and pdf are given as):
Also
10
CDF of exponentially distributed

random variable with = 0.0001
12500
25000
37500
50000
11
Exponential Density Function

(pdf)
12
Memoryless property
Assume X > t, i.e., We have observed that

the component has not failed until time t.
Let Y = X - t , the remaining (residual)

lifetime
GY ( y | t ) = P(Y y | X > t )
= P( X y + t | X > t )
P (t < X y + t )
=
= 1 e y
P( X > t )
13
Memoryless property
Thus GY(y|t) is independent of t and is identical

to the original exponential distribution of X.
The distribution of the remaining life does not
depend on how long the component has been
operating.
Its eventual breakdown is the result of some
suddenly appearing failure, not of gradual
deterioration.
14
Memoryless property
15
Only Continuous Distribution

with Memoryless property
X is a nonnegative R.V. with Memoryless property:
Taking the limit as t approaches zero,
16
Only Continuous Distribution

with Memoryless property
Solution to the differential equation is given by
where K is the const. and
since
and denoting
by constant
Therefore X must have the exponential distribution.

17
Example 3.2
A discrete rv Nt : number of jobs arriving to a

file server in the interval (0, t]
Nt be Poisson distributed (parameter = t )
X : time to next arrival.
Therefore,
X is exponentially distributed with parameter
18
Example 3.3
Web server: time to next request is random

Average rate of requests, = 0.1 reqs/sec.
Number of request arrivals per sec is Poisson
distributed
Or inter-arrival times are EXP(). Therefore,
19
Reliability as a Function of Time
Reliability R(t): prob. that no failure occurs

during the interval (0,t). Let X be the lifetime
of a component subject to failures.
Let N0= total no. of components (fixed); Ns(t)=

surviving ones; Nf(t)= no. failed by time t.
20
Definitions
(Contd.)
Equivalence:
Reliability
Complementary distribution function
Survivor function
R(t) = 1 -F(t)
21
Failure Rate or Hazard Rate
Instantaneous failure rate: h(t)

(#failures/time unit)
As a special case let the rv X be EXP( ). Then the failure

rate is time or age independent:
This is the only continuous distribution with a constant

failure rate (CFR)
22
Hazard Rate and the pdf

f (t )
f (t )
h (t ) =
=
R (t ) 1 F (t )
h(t) t = conditional prob. of system failing in
(t, t + t] given that it has survived until time t.
f(t) t = unconditional prob. of system failing in
(t, t + t].
Analogous to difference between:

probability that someone will die between 90
and 91, given that he lives to 90
probability that someone will die between 90
and 91
23
Reliability from Failure Rate
In the general case, reliability R(t) can be related

to the hazard rate in the following way
Using simple calculus the following applies to any
rv,
24
Failure-Time Distributions
Relationships
f(t)
R(t)
h(t)
R(t)
h(t)
t
f(t)
F(t)
F(t)
f (u ) du
F ' (t )
R ' (t )
1 R (t )
h (t ) e 0
t
h ( u ) du
1 e 0
h ( u ) du
f (u ) du
f (t )
1 F (t )
d
F ' (t )
log e R (t )
dt
f (u ) du (1 F (t ))
e 0
h ( u ) du
1
25
Bathtub curve
DFR phase: Initial design, constant bug fixes
CFR phase: Normal operational phase
IFR phase: Aging behavior
h(t)
(burn-in-period)
(wear-out-phase)
CFR
(useful life)
DFR
IFR
t
Decreasing failure rate
Increasing fail. rate
26
Weibull Distribution
Frequently used to model fatigue failure, ball bearing

failure etc. (very long tails)
F (t ) = 1 e
t 0
Reliability: R(t ) = e
Weibull distribution is capable of modeling DFR ( < 1),
CFR ( = 1) and IFR ( >1) behavior.
is called the shape parameter and is the scale
parameter.
t
27
Weibull Distribution (alternate form)
Some texts use a slightly different form for Weibull:
F (t ) = 1 e
( t )
R(t ) = e
t0
Reliability:
In this text we will use the definition on the previous slide
( t )
28
Example 3.4
Life time X : Weibull distributed with

Observation: 15% components last 90 hrs, but fail
before 100 hrs., i.e.,
Find scale parameter for this Weibull distribution:
29
Failure rate of the Weibull distribution

with various values of and = 1
5.0
1.0
2.0
3.0
4.0
30
Three parameter Weibull Distribution
Sometimes a more complex version of Weibull is used so

that the image of the random variable is in the interval
(,):
31
Infant Mortality Effects in System Modeling
Bathtub curves
Early-life period
Steady-state period
Wear out period
Failure rate models

32
Early-life Period
Also called infant mortality phase or reliability
growth phase or decreasing failure rate (DFR
phase).
Caused by undetected hardware/software
defects that are being fixed resulting in
reliability growth.
Can cause significant prediction errors if
steady-state failure rates are used.
Availability models can be constructed and
solved to include this effect.
DFR Weibull Model can be used.
33
Steady-state Period
Failure rate much lower than in early-life
period.
Either constant (CFR) (age independent)
or slowly varying failure rate.
Failures caused by environmental shocks.
Arrival process of environmental shocks
can be assumed to be a Poisson process.
Hence time between two shocks has
exponential distribution.
34
Wear out Period

Failure rate increases rapidly with age (IFR
phase).
Properly qualified electronic hardware do not
exhibit wear out failure during its intended
service life (as per Motorola).
Applicable for mechanical and other systems.
Again (IFR) Weibull Failure Model can be used
for capturing such behavior.
35
Failure Rate Models
We use a truncated Weibull Model
Failure-Rate Multiplier
7
6
5
4
3
2
1
0
2,190
4,380
6,570 8,760 10,950 13,140 15,330 17,520

Operating Times (hrs)
Infant mortality phase modeled by DFR Weibull

and the steady-state phase by the exponential.
36
Failure Rate Models (cont.)
This model has the form:
1 t 8,760
hW (t ) = C1 t
t > 8,760
= h SS
where:
C1 = hW (1), h SS =
steady-state failure rate

is the Weibull shape parameter
Failure rate multiplier = hW (t ) h SS
37
Failure Rate Models (cont.)
There are several ways to incorporate time dependent

failure rates in availability models.
The easiest way is to approximate a continuous
function by a decreasing step function.
Failure-Rate Multiplier
7
6
h1
5
4
3
h2
h SS
1
0
2,190
4,380
6,570 8,760 10,950 13,140 15,330 17,520

Operating Times (hrs)
38
Failure Rate Models (contd.)

Here the discrete failure-rate model is
defined by:
0 t < 4,380
hW (t ) = h1
= h2
4,380 t < 8,760
= h ss
t 8,760
The approximation can be improved by

taking smaller time steps.
39
HypoExponential (HYPO)
HypoExp: multiple Exp stages in series.

2-stage HypoExp denoted as HYPO(1, 2). The
density, distribution and hazard rate function are:
HypoExp is an IFR as its h(t): 0 min{1, 2}

Disk service time may be modeled as a 3-stage
Hypoexponential as the overall time is the sum of
the seek, the latency and the transfer time.
40
HypoExponential pdf and CDF
Hypo(1,5)
Hypo(1,5) pdf
Hypo(1,5) CDF
41
HypoExponential used in software

rejuvenation models
Preventive maintenance is useful only if failure rate is increasing
A simple and useful model of increasing failure rate:
Robust state
Failure
probable
state
Failed state
Time to failure: Hypo-exponential distribution

Increasing failure rate
aging
42
Erlang Distribution
Special case of HYPO: All stages have same rate.
[X > t] = [Nt < r] (Nt : no. of stresses applied in (0,t]

and Nt is Poisson (parameter: t). This interpretation
gives,
43
Erlang Distribution
If we set the parameter r=1, we get the

exponential distribution
Erlang distribution can be used to
approximate the deterministic variable,
since if the mean is kept same but
number of stages are increased, the pdf
approaches the delta (impulse) function
in the limit.
44
Erlang density function
If we vary r keeping r/ constant, pdf of r-stage Erlang approaches

an impulse function at r/ .
45
Erlang Cumulative Distribution Function
And the cdf approaches a step function at r/. In other words

r-stage Erlang can approximate a deterministic variable.
46
Gamma Random Variable
A basic distribution of statistics for non-negative

variables (see Section 3.9 and Chapter 10)
Gives distribution of time required for exactly r
independent events to occur, assuming events take
place at a constant rate (p. 131 of text). Used frequently
in queuing theory, reliability theory
Example: Distribution of time between re-calibrations of
instrument that needs re-calibration after r uses; time
between inventory restocking, time to failure for a
system with cold standby redundancy (Ex. 3.25)
Erlang, exponential, and chi- square distributions are
special cases.
47
Gamma Random Variable
Gamma density function is,
() = (-1) (-1); (1/2)=

Because (1)=1, it follows that (r)=(r-1) (r-1)=...=(r-1)! So
gamma with an integer valued shape parameter is the Erlang
distribution
Gamma with shape parameter = and scale parameter
= n/2 is known as the chi-square random variable with n degrees
of freedom.
48
Gamma distribution: failure rate

Gamma distribution can capture all three types
failure rate behavior, viz. DFR, CFR or IFR

depending on the value of the shape paramter
= 1: CFR
<1 : DFR
>1 : IFR
49
Gamma density function
50
HyperExponential Distribution (HyperExp)
Hypo or Erlang have sequential Exp( ) stages.

When there are alternate Exp( ) stages it becomes
Hyperexponential.
CPU service time may be modeled by HyperExp.

In workload based software rejuvenation model we found
the sojourn times in many workload states have this kind of
distribution.
51
Hyper Exponential Vs. Exponential CDF
Distribution of measured CPU service time

may be better described by the HyperExp( )
as compared to the EXP( ) distribution.
52
Log-logistic Distribution
Log-logistic can model more complex failure rate

behavior than simple CFR, IFR, DFR cases.
For, > 1, the failure rate first increases with t ;

after momentarily leveling off, it decreases with time.
This is known as the inverse bath tub shape
curve.
Useful in modeling software reliability growth .
53
Log-logistic failure rate
54
Gaussian (Normal) Random Variable
A basic distribution of statistics. Many applications arise from

central limit theorem (average of values of n observations
approaches normal distribution, irrespective of form of original
distribution under quite general conditions).
Consequently, appropriate model for many, but not all, physical
phenomena.
Example: Distribution of physical measurements on living
organisms, intelligence test scores, product dimensions, average
temperatures, and so on.
Many methods of statistical analysis presume normal
distribution.
In a normal distribution, about 68% of the values are within one
standard deviation of the mean and about 95% of the values
are within two standard deviations of the mean.
55
Gaussian (Normal) Random Variable
Bell shaped and symmetrical pdf intuitively pleasing!

Central Limit Theorem: sum of a large number of
mutually independent rvs (having arbitrary distributions)

starts following Normal distribution as n
: mean, : std. deviation, 2: variance (N(, 2))

and completely describe the rv. This is significant in
statistical estimation/signal processing/communication
theory etc.
Mean, median and mode are all equal; infinite range
56
Normal Density with

parameter =2 and =1
57
Normal Distribution (contd.)
Failure rate h(t) follows IFR behavior.
Hence, normal distribution is suitable for

modeling long-term wear or aging related
failure phenomena.
See page 138-140 for Examples.
58
Failure rate for Normal

distribution
h(t) for normal distribution is IFR
59
Infinitely many normal pdfs
By changing the two parameters, we can

get infinitely many normal densities
60
Normal Distribution (contd.)
No closed form for the CDF; how do we

determine P(a< X <b)?
Answer: use tables after a transformation to
standard normal
N(0,1) is called standard normal distribution.
X ~ N(, 2)) then Z=(X- )/ is N(0,1)
N(0,1) is symmetric i.e.
f(x)=f(-x)
F(-z) = 1-F(z).
61
Example 3.5
X: amplitude of an analog signal at a detector

X has a normal distribution N(200,256)
Find P(X > 240)
62
Example 3.5 (contd.)
Find P(X>240 |X>210)
63
Example 3.6
X: Wearout phase lifetime of a subsystem is normal

N(105,106) (in hour units)
Find R9,000(500) and R11,000(500)
64
Similarly,
65
Uniform Random Variable
Unif(a,b) pdf constant over the interval
(a,b) and CDF is the ramp function
All (pseudo) random number generators

generate random deviates of Unif(0,1)
distribution; that is, if a large number of
random variables are generated and their
empirical distribution function is plotted, it will
approach this distribution in the limit.
66
Uniform density function

Uniform pdf Unif(a, b)
67
Uniform distribution
The distribution function is given by:

0,
x a
F ( x) =
,
b a
1,
x<a
a x<b
bx
68
Uniform CDF
69
Erlang to approximate Uniform
Uniform random variable is sometimes

approximated by a Erlang random
variable
We will see an example of this in
Chapter 8
In the next two slides, pdfs and CDFs of
Unif(0,1) and 3-stage Erlang with
parameter =6 are compared
70
Comparison of
probability density functions (pdf)
1.8
1.6
1.4
1.2
1
pdf
3-stage Erlang pdf

U(0,1) pdf
0.8
0.6
0.4
0.2
1.
02
0.
96
0.
9
0.
84
0.
78
0.
72
0.
66
0.
6
0.
54
0.
48
0.
42
0.
36
0.
3
0.
24
0.
18
0.
12
0.
06
time
71
Comparison of cumulative
distribution functions (cdf)
1.2
3-stage Erlang cdf
0.6
U(0,1) cdf
0.4
0.2
1.
02
0.
9
0.
96
0.
84
0.
78
0.
72
0.
6
0.
66
0.
48
0.
54
0.
42
0.
3
0.
36
0.
24
0.
18
0.
12
0
0
0.
06
cdf
0.8
time
72
Pareto Distribution
Also known as the power law or longtailed distribution also, heavy-tailed

distribution.
Found to be useful in modeling of
CPU time consumed by a request.

Web file size on internet servers.
Number of data bytes in FTP bursts.
Thinking time of a Web browser.
73
Pareto Distribution
The density is given by
f ( x) = k x
(Contd.)
, x k, , k > 0
The Distribution is given by
And the failure rate is given by
74
The CDF of Pareto Distribution
75
The pdf of Pareto Distribution
76
Log-normal
Permits representation of random variable whose logarithm

follows normal distribution. Model for a process arising from
many small multiplicative errors. Appropriate when the value of
an observed variable is a random proportion of the previously
observed value.
In the case where the data are log-normally distributed, the
geometric mean acts as a better data descriptor than the
mean.The more closely the data follow a lognormal distribution,
the closer the geometric mean is to the median
Example: Repair time distribution; life distribution of some
transistor types.
pdf is given by:
77
Defective Distribution
Recall that
Example:
If p c < 1, then , FX ( x ) = p c (1 e
) is a
defective exponentia l distributi on .
This defect (also known as the mass at infinity)

could represent the probability that the program
will not terminate (1-pc). Continuous part can
model completion time of program; we will see
many examples in later chapters.
There can also be a mass at origin.
78
The CDF of a Defective random

variable
79
Functions of Random Variables
Often, rvs need to be

transformed/operated upon.
Y = (X) : so, what is the distribution or

the density of Y ?
Example: Y = X2
Example: Y= --1ln(1-X)
80
Example 3.8
Distribution for Y = (X) = X2
81
Example 3.9
In Example 3.8, assume X to be N(0,1):
Using result from Example 3.8:
Which is also known as chi-square distribution with 1

degree of freedom
82
Example 3.10
Let X be uniformly distributed, Unif(0,1)

Then, if Y = --1 ln(1-X) is EXP().
This transformation is used to generate a random

variate (or deviate) of the Exp() distribution
83
Theorem: pdf for a transformed RV
84
Random Variate Generation
Random variate is defined as a typical value

sampled from a given distribution. If we take
a large number (ideally infinite) of them and
plot a histogram, it will approach the original
pmf or pdf.
Goal: Study methods of generating random
deviates of a given distribution, assuming a
routine to generate uniformly distributed
random numbers is available.
Note that distribution of interest can be
discrete or continuous.
85
Some generation Methods
Some popular methods of generating random

variate are:
Inverse Transform Method

Convolution Method
Direct Transformation of Normal Distribution.
Acceptance-Rejection Method
86
Inverse Transform
Based on the following idea:

If F(x) is strictly monotonic distribution function
and U is uniformly distributed over the interval
(0, 1).
Then the new random variable X=F-1(U) has the
the CDF F.
Method:
A random number u from a uniform distribution

over (0, 1) is generated and then the F is
inverted to generate random deviate x of X.
F-1(u) gives the required value of x.
87
Example 3.11
Generate random variate x with distribution F=FX

Let, Y = F(X)
FY(u) = FY(F 1(u)) = u, or, Y=F(X) has pdf,
Hence, to generate random variate (deviate) with

distribution F,
1. Generate random number u
-1
2. Find x = F (u) and x will be a random deviate with distribution F
3. If x = -1 ln(1-u), then x will be a random deviate of EXP()
distribution.
Inversion can be done in closed form, graphically or

using a table
88
Inverse Transform
Variates of Exponential, Uniform, Weibull,

Pareto, Rayleigh, Triangular, Log-logistic and
many others can be generated by this
method.
Variates of empirical and discrete
distributions like Bernoulli and Geometric can
also be generated using this idea.
It is most useful when the inverse of the
CDF, F(.) can be easily computed in closed
form although a numerical or tabular method
can also be used.
89
Some Examples
CDF
F ( x) = 1 e x , x 0
where is its failure rate (1/ is the mean).
Random Variate x
ln(1 u )
x=
where u is drawn from uniform distribution Unif(0,1).

Since (1-u) is also a random number , use the
simpler formula
x=
ln(u )
90
Exponential variate
Next we will show that if enough variates are

sampled then the sample of generated
numbers is sufficient to describe the pdf of
the distribution.
We see that as we increase number of
observations the plot becomes closer and
f (t ) = e for t 0
closer to the theoretical pdf
pdf of exponential distribution with mean 1 is
plotted
Three cases are taken with 10, 100 and
1000 observations.
91
t
Theoretical Exp pdf
For 10 Observations
1.2
1.2
0.8
0.8
0.6
0.6
0.4
0.4
0.2
0.2
2 3
4 5 6
7 8
9 10 11 12 13 14 15 16
For 100 Observations

1.2
10 11 12 13 14 15 16
For 1000 observations

1.2
0.8
0.8
0.6
0.6
0.4
0.4
0.2
0.2
11
13
15
2 3
4 5 6
7 8
9 10 11 12 13 14 15 16
92
Some Examples: Example 3.12
Weibull Distribution
CDF
FX ( x ) = 1 e
where is the scale parameter and is the shape parameter.

1
Random Variate x
ln(1 u )
x=

Simplified formula
1
ln(u )
x =
93
Extensions to Example 3.12
Write down the random deviate formual for the alternate for
of the Weibull Distribution
CDF
FX ( x) = 1 e
( x )
And for the three parameter Weibull distribution:
CDF
94
Some Examples: Pareto
Pareto Distribution
CDF
FX ( x ) = 1
xk
x
=0
xk
where k>0 is location parameter and is shape
parameter.
Random Variate x
k
x=
(1 u )

k
Simplified Formula
x= 1
95
Some Examples: Rayleigh
Rayleigh Distribution
x 2 / 2 2
x0
CDF FX ( x ) = 1 e
=0
otherwise
where 2 is the variance.
Random Variate x
x = 2 2 ln(1 u )
where u is drawn from uniform distribution

Unif(0,1).
x = 2 2 ln(u )
Simplified Formula
96
Some Examples: Log-logistic
Log-Logistic Distribution
CDF
FX ( x) = 1
1
1 + ( x )
where >0 is the scale parameter and >0 is shape

parameter.
1
Random Variate x
1 u
x=
u
1
where u is drawn from uniform distribution

Unif(0,1).
97
Random variate Table

Name
Density
Expo
e x x > 0
f(x)
1 e x x > 0
1 e x
Weibull
e x x > 0
Pareto
k x 1 x > k ,
,k > 0
Rayleigh
1 x 2
x 2 / 2 2
exp
e
2
2
x
X=F-1(u) Simpl.
F(x)
k
1
x
x=
form
ln(1 u )
x=
ln(1 u )
x=
ln(u )
1
1
ln(u )
x =
1
k
x= 1
u
2
x 0 x = 2 ln(1 u )
x = 2 2 ln(u )
x=
k
(1 u )
98
Random variate Table

F(x)
Name
Density
LogLogistic
Cauchy
1
(t )
1
t0
1 + ( t )
[1 + (t ) ]2
Triangul
ar
f(x)
(x + )
2
2 x
1
a a
1 1
x
+ arctan
2

2
x2
x
a
2a
X=F-1(u) Simplifie
d form
1 u
x=
1 u
x=
1 u
1 u
tan (u )
tan (u )
a (1 1 u )
a (1 u )
1
2
99
For Discrete distributions
We want to generate X having pmf

p j = P( X = x j ), j = 0,1,...
j
and distribution F ( x j ) = pi
i =0
If u is the uniform random number then
This is as
x = x j , if F ( x j 1 ) < u F ( x j )
j 1
i =1
i =1
P( X = x j ) = P ( pi < U pi ) = p j
100
For discrete distributions
Bernoulli distribution
CDF with parameter (1-q)
FX ( x ) = 0 x < 0
= q 0 x <1
=1 x 1
Inverse function is given by

F 1 (u ) = 0 0 < u q
=1 q < u 1
By generating u from Unif(0,1) function, we can

get Bernoulli random variate from above.
101
For Geometric Distributions
p (1 p ) j 1 j 1
Geometric pmf is given by

F ( j ) = 1 (1
Distribution is given by
So geometric random variate j satisfies
p) j
1 (1 p ) j 1 < u 1 (1 p ) j
(1 p ) j 1 < 1 u (1 p ) j
Hence Geometric x = min{ i : (1 p ) i < 1 u }

random variate x
= min{ i : i >
log( 1 u )
}
log( 1 p )
log( 1 u )
=
log(
1
p
)
Since (1-u) is also uniform random number it becomes

log(u )
log(1 p )
102
Example 3.13
Scaling a random variable X

Let fX(x) = e-x and Y = rX
Hence, Y is also EXP( ) with parameter /r

Thus the exp distribution is closed under a
multiplication by a scalar
103
Example 3.14
Distribution for Y = (X) = eX, given X is N(, 2)
Random variable Y is said to have a log-normal

distribution.
Repair times are often found to follow this
distribution
104
Jointly Distributed RVs
Joint Distribution Function:
Independent rvs: iff the following holds:
Independent rvs: iff the following holds:

105
Joint Distribution Properties
106
Joint Distribution Properties (contd.)
107
Order statistics, k of n, TMR
108
Order Statistics: k of n
X1 ,X2 ,..., Xn iid (independent and identically

distributed) random variables with a common
distribution function F and common density f.
Let Y1 ,Y2 ,...,Yn be random variables obtained by

permuting the set X1 ,X2 ,..., Xn so as to be in
increasing order.
To be specific:
Y1 = min{X1 ,X2 ,..., Xn} and

Yn = max{X1 ,X2 ,..., Xn}
109
(Continued)
The random variable Yk is called the kth

ORDER STATISTIC.
If Xi is the lifetime of the ith component in a
system of n components. Then:
Y1 will be the overall series system lifetime.

Yn will denote the lifetime of a parallel system.
Yn-k+1 will be the lifetime of an k-of-n system.
110
(Continued)
To derive the distribution function of Yk, we note that

the probability that exactly j of the Xi's lie in (- ,y]
and (n-j) lie in (y, ) is (n Bernoulli trials; p=F(y)):
n j
F ( y ) [1 F ( y )]n j
j
< y <
hence
n j
FYk ( y ) = F ( y ) [1 F ( y )]n j , < y <
j =k j
n
111
Overview:
General iid Random Variables
Let Y1 ,,..., Yn denote the order statistics of the random variables X1 ,..., Xn,,
which are iid with common distribution function F.
Y1
Yn k +1
FYn k +1 ( y )
n
n j
= F ( y )[1 F ( y )]n j
j = n k +1 j
Rk |n (t ) = 1 FYn k +1 (t )
n
n j
= R (t )[1 R(t )]n j
j =k j
Yn
112
Overview:
Y1
Yn k +1
Yn
FYn k +1 ( y )
n
n j
= F ( y )[1 F ( y )]n j
j = n k +1 j
FYn ( y ) = [ F ( y )]
Rk |n (t ) = 1 FYn k +1 (t )
n
n j
= R (t )[1 R(t )]n j
j =k j
R parallel (t ) = 1 FYn (t )
= 1 [ F (t )]n = 1 [1 R(t )]n
113
Overview:
Y1
Yn k +1
Yn
FY1 ( y ) = 1 [1 F ( y )]
FYn k +1 ( y )
n
n j
= F ( y )[1 F ( y )]n j
j = n k +1 j
FYn ( y ) = [ F ( y )]
Rseries (t ) = 1 FY1 (t )
= [1 F (t )]n = [ R(t )]n
Rk |n (t ) = 1 FYn k +1 (t )
n
n j
= R (t )[1 R(t )]n j
j =k j
= 1 [ F (t )]n = 1 [1 R(t )]n
114
Applications of order statistics
Reliability of a k out of n system

n
Rkofn (t ) = ( nj )[ R (t )] j [1 R(t )]n j

j =k
Series system:
Rseries (t ) = [ R(t )]
or Ri (t )
i =1
Parallel system:
R parallel (t ) = 1 [1 R(t )] or1 (1 Ri (t ))

n
i =1
Minimum of n EXP random variables is special case of Y1 =

min{X1,,Xn} where Xi~EXP(i) so Y1~EXP( i)
Thus the exponential is closed under series composition but not
under the parallel composition .
115
Example 3.16
Series system lifetime distribution

ith components life time distribution ~ EXP(i)
Lifetime distribution of series system of

components with each component having
EXP( ) distribution is also EXP(s) with
116
Example 3.17 (Parts count

method)
Assuming that times to failure of all chip types are

exponentially distributed with the following failure
rates:
117
Example 3.18
Hence series system of Example 3.17 has a constant

failure rate.
What about a parallel system of n such components:
118
Example 3.20
Arrivals from n sources : si generates Ni(t) tasks in

time t . Ni (t) Poisson with parameter it
Xi :time between two successive arrivals from si has
EXP(i) distribution.
Total no. of jobs
is also Poisson with rate parameter
The jobs arrive in the pooled stream with interarrival
time,
119
Triple Modular Redundancy (TMR)

R(t)
R(t)
Voter
R(t)
An interesting case of order statistics occurs when we

consider the Triple Modular Redundant (TMR) system (n = 3
and k = 2). Y2 then denotes the time until the second
component fails. We get:
RTMR (t ) = 3R (t ) 2 R (t )
2
120
TMR
(Continued)
Assuming that the reliability of a single

component is given by,
R (t ) = e
we get:
RTMR (t ) = 3e
2 t
2e
3t
121
TMR
(Continued)
In the following figure, we have plotted RTMR(t)

vs. t as well as R(t) vs. t.
Also graphs have been plotted for comparison
between TMR and TMR/Simplex using SHARPE
GUI. A step by step procedure has been shown.
We see that TMR improves reliability over the
simplex for short mission times (defined by
t < ln 2/); for longer mission times, TMR has
lower reliability than simplex
122
TMR
(Continued)
123
124
125
126
127
128
129
130
131
132
133
134
Workstations & File server (WFS)

Example RBD Approach
135
The WFS Example
Computing system consisting of:
A file-server
Two workstations
Computing network connecting them
136
RBD for the WFS Example

Workstation 1
File Server
Workstation 2
137
RBD for the WFS Example

Rw(t): workstation reliability
Rf (t): file-server reliability
System reliability R(t) is given by:
R(t) = [1 - (1 - Rw(t))2] Rf (t)

Note: applies to any time-to-failure distributions
138
Snapshot of the GUI
139
bind
lambdaW 0.0001
lambdaF 0.0003
end
block wfs1
* each component is non-restorable and has exp time to fail dist
comp Workstation exp(lambdaW)
comp FileServer exp(lambdaF)
parallel work Workstation Workstation
series sys work FileServer
end
* define function R(t) for reliability at time t
func R(t) 1-tvalue(t;wfs1)
* vary the time t from t=0 to 10000 in steps of 1000 hours and print R(t)
loop t,0,10000,1000
expr R(t)
end
140
end
R(t) vs. time

exp
1.2
1
Reliability
0.8
0.6
exp
0.4
0.2
0
0
1000
2000
3000
4000
5000
6000
7000
8000
9000
10000
tim e
141
Example 3.21
A system with n workstations and m file servers
System Operational: k workstation & j file servers

WS reliability is Rw(t) and FS reliability is Rf(t)
Assuming all devices fail independently,
142
2 Control and 3 Voice Channels

Example
voice
control
voice
control
voice
143
Description
Each control channel has a reliability Rc(t)
Each voice channel has a reliability Rv(t)
System is up if at least one control channel and

at least 1 voice channel are up.
Reliability:
R (t ) = [1 (1 Rc (t )) ][1 (1 Rv (t )) ]
2
144
Reliability block diagram

model in SHARPE GUI
145
Define the components in the

block diagram model
146
Output selection
147
Results from SHARPE
148
Plot definition
149
Reliability vs. time
150
Definition of another plot
151
Reliability vs. lambda
152
2c3v as a Fault Tree

or
Structure Function:
and
and
c1
c2
v1
= c1 c2 + v1 v2 v3
v2 v3
2 Control and 3 Voice Channels Example

153
Fault Tree Example (contd.)
Reliability of the system:
Assume Rc (t ) = e
c t
and Rv (t ) = e
v t
R(t) = [1 (1 Rc (t)) ][1 (1 Rv (t)) ]

2
ct
= (2e
2ct
vt
)(3e
2vt
3e
3vt
+e
154
Fault tree input in SHARPE GUI
155
Example 3.22
Shuffle exchange network (SEN) with N = 2n inputs
(N/2) switching elements/stage; log2N such stages
Single switch element

8 x 8 SEN
156
We are interested in finding the reliability RSEN(t) of

this SEN given individual switch reliability rSE(t)
157
Example 3.23
SEN+ has an extra stage of N/2 switching elements

to increase reliability; 00 01 has two paths.
rSE(t): time-dependent reliability of an SE
SEN: (N/2) log2N = 4 elements

158
Example 3.23 (contd)
SEN+ reliability in this case is worse that SEN

For 8X8 case, SEN+ reliability works out to be better
than SEN
8x8 SEN+
159
Overview: Exponential iid

Random Variables
which are iid with common distribution function F(x)=1 exp(-x).
Y1
FY1 ( y ) = 1 [exp( y )]n Rseries (t ) = 1 FY (t )

1
= 1 exp( ny )
= exp( nt )
Yn k +1
FYn k +1 ( y )
n
n
= [1 exp( y )] j
j = n k +1 j
exp( ( n j )y )
Rk |n (t ) = 1 FYn k +1 (t )
Yn
FYn ( y ) = [1 exp( y )]
Y1 EXP( n )
= 1 [1 exp( t )]n
160
Overview: Exponential iid

Random Variables
which are iid with common distribution function F(x)=1 exp(-x).
Y1
FY1 ( y ) = 1 [exp( y )]n Rseries (t ) = 1 FY (t )

1
= 1 exp( ny )
= exp( nt )
Y1 EXP( n )
Yn k +1
FYn k +1 ( y )
n
n
= [1 exp( y )] j
j = n k +1 j
exp( ( n j )y )
Rk |n (t ) = 1 FYn k +1 (t )
Yn
FYn ( y ) = [1 exp( y )]
R parallel (t ) = 1 FYn (t ) Y HYPO( n,

n
( n 1),..., )
= 1 [1 exp( t )]n
Yn k +1
HYPO( n,
( n 1),..., k )
161
Overview: Exponential
Independent Random Variables
Let Y1 ,,..., Yn denote the order statistics of the independent random
variables X1 ,..., Xn,. The distribution function of Xi is F(xi)=1 exp(ix).
n
Y1
FY1 ( y ) = 1 exp( i y ) R (t ) = 1 F (t )
series
Y1
i =1
= 1 exp( y i =1 i )
Yn k +1
Complicated
n
Yn
FYn ( y ) = Fi ( y )
n
i =1
= [1 exp( i y )]
i =1
= exp( t i =1 i )
n
Complicated
n
= 1 [1 exp( i t )]
i =1
Y1 EXP( i =1 i )
n
Complicated
Complicated
162
Sum of Random Variables
Z = (X, Y)
For the special case, Z = X + Y

The resulting pdf is (assuming independence),
Convolution integral (modify for the non-negative
case)
163
Convolution (non-negative case)

Z = X + Y, X & Y are independent random
variables (in this case, non-negative)
t
f Z (t ) = f X ( x) fY (t x) dx
0
The above integral is often called the

convolution of fX and fY. Thus the density of
the sum of two non-negative independent,
continuous random variables is the convolution
of the individual densities.
164
Example 3.24: Multithreaded program

performance
Three independent computer tasks 1, 2 and 3

Precedence relationship: 3 has to wait for both 1
and 2 to complete
T1, T2 and T3 : respective random execution times
Total execution time = max{T1, T2 }+ T3 =M + T3
T1 and T2 ~Unif(t1- t0 , t1+ t0); T3 ~Unif(t3- t0 , t3+ t0)
Find Probability that T > t1+ t3
165
The pdfs are:
166
167
We need to find P(T > t1+ t3); A : shaded area
168
Reliability Modeling Examples
Sums of exponential random variables

appear naturally in reliability modeling
Cold-standby redundancy
Warm-standby redundancy
Hot-standby redundancy
Triple Modular Redundancy (TMR)
TMR/Simplex
k-out-of-n Redundancy
169
Two component system
With respective lifetime random

variable, X and Y, assumed independent
Series system (Z=min{X,Y})
Parallel System (Z=max{X,Y})
Cold standby: the lifetime Z=X+Y
170
Cold standby (standby redundancy)

X
Lifetime of
Active
Lifetime of
Spare
EXP()
EXP()
Total lifetime 2-Stage Erlang
R(t ) = (1 + t )e
Assumptions (to be relaxed later):

Detection & Switching perfect;
Spare does not fail.
EXP()
EXP()
171
Cold standby derivation
X and Y are both EXP() and

independent.
Then
f Z (t ) = e e
x
( t x )
dx
2 t
= e
dx
0
= te , t > 0
2
172
Cold standby derivation
(Continued)
Z is two-stage Erlang Distributed

t
FZ (t ) = f Z ( z )dz = 1 (1 + t )e
R(t ) = 1 F (t )
t
= (1 + t )e , t 0
173
Convolution: r-stage Erlang
The general case of r-stage Erlang Distribution
When r sequential phases have independent

identical exponential distributions, then the resulting
random variable is known as r-stage (or r-phase)
Erlang and is given by:
174
Convolution: Erlang
EXP()
EXP()
f (t ) =
(Continued)
EXP()
r r 1 t
t e
(r 1)!
(t ) k t
F (t ) = 1
e
k!
k =0
r 1
175
Standby Sparing Example 3.26
System with n processors whose lifetimes are iid, following an

exponential distribution with failure rate .
Two modes
Only 1 of n is active, others are cold standby
All n are active, working in parallel
We can see that:
since
(t ) t
t n
e
(
1
e
)
k!
k =0
n 1
However, parallel arrangement delivers more capacity.

176
Derivation of the result
Let X1, X2, , Xn be the times to failure random

variables of the n processors
At time Y1=min {X1, X2, , Xn}, one processor has
failed and remaining (n-1) are working
Computing capacity will also drop to (n-1),
177
Derivation of the result (contd)
From the diagram, Cn is the area under the curve and

we wish to find distribution for Cn.
First find distribution for Yj+1-Yj
Assume that all processor lifetimes are EXP(), then
we assert, (Yj+1-Yj)~EXP[(n-j) ].
Assume Y0 = 0,
(Y1-Y0)=Y1=min{X1, X2, , Xn} ~EXP(n)

Hence, assertion is true for j=0.
After j procs have failed, the residual lifetimes are W1,
W2, , Wn-j , each of which is EXP() due to the
memoryless property of the exponential distribution.
178
Derivation of the result (contd2)
(Yj+1-Yj) is then given by,

(Yj+1-Yj) = min{W1, W2, , Wn-j}
(Yj+1-Yj) ~ EXP[(n-j) ] using result of Example 3.16

Using the result of Example 3.13,
(n-j) (Yj+1-Yj) ~ EXP().

Therefore, Cn is the sum of n independent identically
distributed exponential rvs or Cn is n-stage Erlang.
Thus the total computing capacity delivered before
failure has the same distribution in both the modes of

operation.
179
Warm standby
With Warm spare, we have:
Active unit time-to-failure: EXP()
Spare unit time-to-failure: EXP()
EXP(+ )
EXP()
2-stage hypoexponential distribution

180
Warm standby derivation
First event to occur is that either the active or

the spare will fail. Time to this event is
min{EXP(),EXP()} which is EXP( + ).
Then due to the memoryless property of the
exponential, remaining lifetime is still EXP().
Hence system lifetime has a two-stage
hypoexponential distribution with parameters
1 = + and 2 = .
181
Warm standby derivation
(Continued)
X is EXP(1) and Y is EXP(2) with 1 2;

X and Y are independent.
t
Then f Z ( t ) = 1e 1 x 2 e 2 ( t x ) dx
1 2 2t
1 2 1t
e
e .
=
+
1 2
2 1
This is the density of the 2-stage hypoexponential distribution with parameters 1
and 2.
182
Hot standby (Active/Active)

With hot spare, we have:
Active unit time-to-failure: EXP()
Spare unit time-to-failure: EXP()
EXP(2)
EXP()
2-stage hypoexponential
183
TMR and TMR/simplex

as hypoexponentials
EXP(3)
EXP()
TMR/Simplex
EXP(3)
EXP(2)
TMR
184
Hypoexponential: general case

r
X , where X1 ,X2 , , Xr are mutually independent
Z=
i =1
and Xi is exponentially distributed with parameter i

where
i j for i j
Then Z is a r-stage hypoexponentially distributed

random variable.
EXP(1)
EXP(2)
EXP(r)
185
Hypoexponential: general case
Density function:
See Page 174, Theorem 3.4 of the Text.
186
k of n system lifetime, as a
hypoexponential
At least, k out of n units should be operational
for the system to be up. Here failure rate of
each unit is .
EXP(n)
Y1
EXP((n-1))
Y2
...
EXP(k)
Yn-k+1
EXP((k-1))
Yn-k+2
...
EXP()
Yn
187
k of n with warm spares
At least, k out of n + s units should be operational for the

system to be up. Initially n units are active and s units are
warm spares. The failure rate of a unit when active is
and the failure rate of a unit when spare is .
EXP(n
+s)
EXP(n
+(s-1))
...
EXP(n
+ )
EXP(n)
...
EXP(k)
188
Sums of Normal Random Variables
X1, X2, .., Xn are mutually independent normal

rvs, then, the rv Z = (X1+ X2+ ..+Xn) is also
normal with
The sum of mutually independent normal

random variables is also normal.
X1,..., Xn are independent standard normal. Then
ffff
follows the gamma distribution
or the 2 distribution with n degrees of freedom.
189
Example 3.34
A sequence of independent, identically distributed

random variables, X1,X2, . . . , Xn, is known as a
random sample of size n.
In many problems of statistical sampling theory, it is
reasonable to assume that the underlying distribution
is the normal distribution.
Thus let
Then from last slide, we obtain
190
191
Example 3.35
are independent, and
Variables to polar co-ordinates
Y = X12 + X22 is exp distributed with parameter 0.5

192
Example 3.36
Assume that X1,X2, . . . , Xn are mutually independent

identically distributed normal random variables such that
Then
has the standard normal distribution.
Therefore,
has the the 2 distribution with n degrees of freedom.

The random variable
may be used as an estimator of the variance

.
193
Example 3.37
However, the mean of the population is often

unknown.
can then be estimated from the sample variance
It can be shown that the random variable
has the 2 distribution with n1 degrees of freedom.

194
Example 3.39
Assume that X1,X2, . . . , Xn are mutually independent

identically distributed normal random variables such that
Then
Also
Therefore,
has the standard normal distribution.

has the
distribution.
195

Chap3f Secure PDF

Transféré par

Informations du document

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Chap3f Secure PDF

Transféré par

Droits d'auteur :

Formats disponibles

Probability and Statistics with

Reliability, Queuing and Computer

Chapter 3: Continuous Random Variables

Copyright 2006 by K.S. Trivedi

We will also allow defective distributions. Defective

Unless otherwise specified, we will assume all distributions

Copyright 2006 by K.S. Trivedi

CDF (Cumulative Distribution Function)

Probability Distribution Function (PDF)

FX (x) or FX(t) or F(t)

probability density function (pdf)

X : continuous rv, then,

CDF and pdf can be derived from each other

Copyright 2006 by K.S. Trivedi

probability density function

Copyright 2006 by K.S. Trivedi

Random variable X : time (years) to complete a project

fX clearly satisfies property (f1).

Prob. of completing project in less than 4 months,

Copyright 2006 by K.S. Trivedi

Arises commonly in reliability & queuing theory.

Copyright 2006 by K.S. Trivedi

The use of exponential distribution is an assumption

Copyright 2006 by K.S. Trivedi

Mathematically (CDF and pdf are given as):

Copyright 2006 by K.S. Trivedi

CDF of exponentially distributed

Copyright 2006 by K.S. Trivedi

Exponential Density Function

Copyright 2006 by K.S. Trivedi

Assume X > t, i.e., We have observed that

Let Y = X - t , the remaining (residual)

Thus GY(y|t) is independent of t and is identical

Copyright 2006 by K.S. Trivedi

Only Continuous Distribution

Taking the limit as t approaches zero,

Copyright 2006 by K.S. Trivedi

Only Continuous Distribution

Therefore X must have the exponential distribution.

A discrete rv Nt : number of jobs arriving to a

X is exponentially distributed with parameter

Copyright 2006 by K.S. Trivedi

Web server: time to next request is random

Or inter-arrival times are EXP(). Therefore,

Copyright 2006 by K.S. Trivedi

Reliability as a Function of Time

Reliability R(t): prob. that no failure occurs

Let N0= total no. of components (fixed); Ns(t)=

Copyright 2006 by K.S. Trivedi

Complementary distribution function

Failure Rate or Hazard Rate

Instantaneous failure rate: h(t)

As a special case let the rv X be EXP( ). Then the failure

This is the only continuous distribution with a constant

Hazard Rate and the pdf

Analogous to difference between:

Reliability from Failure Rate

In the general case, reliability R(t) can be related

Copyright 2006 by K.S. Trivedi

Increasing fail. rate

Copyright 2006 by K.S. Trivedi

Frequently used to model fatigue failure, ball bearing

Copyright 2006 by K.S. Trivedi