TEST BANK FOR The Art of Computer Systems Performance By Raj Jain

􀀀 Compare the ratio with system A as the base
System Workload 􀀀 Workload Average
A 􀀀 􀀀 􀀀
B 􀀀
Considering the ratio of performance with system A as base we con
clude that system B is better
􀀀 Compare the ratio with system B as the base
System Workload 􀀀 Workload Average
A 􀀀
B 􀀀 􀀀 􀀀
Considering the ratio of performance with system B as base we con
clude that system A is better

􀀀 incomplete
Can be done

􀀀 a Measurements Run your favourite programs and pick the one that
runs them faster
b Use measurements and simulations of various network con    guirations
c Measurement
d 
a Analytical modelling

b Analytical modelling and simulations

c Extensive simulations and modelling
a 􀀀 Response time for commonly used programs
􀀀 Failure rate 
rate of crashing
􀀀 Storage capacity
􀀀 Userfriendliness
b 􀀀 Query response time
􀀀 Failure rate
􀀀 Storage capacity
􀀀 Usability
c 􀀀 Capacity
􀀀 Response time
􀀀 Failure rate
d 􀀀 Response time

􀀀 The following information is from SPEC 
Standard Performance Evaluation
Corporation home page
CPU benchmarks
CINT current release  Rel 􀀀􀀀 Integer benchmarks contains
Name Application
espresso Logic Design
li Interpreter
eqntott Logic Design
compress Data Compression
sc Spreadsheet
gcc Compiler
CFP current release  Rel 􀀀􀀀
Floating point benchmark suite contains
Name Application
spiceg Circuit Design
doduc Simulation
mdljdp Quantum Chemistry
wave Electromagnetism
tomcatv Geometric Translation
ora Optics
alvinn Robotics
ear Medical Simulation
mdljsp Quantum Chemistry
swm Simulation
sucor Quantum Physics
hydrod Astrophysics
nasa NASA Kernels
fpppp Quantum Chemistry
More information about these benchmarks can be found in http􀀀performancenetliborgperforma
web page
A C program to implement sieve workload
􀀀
seivec Program to implement sieve workload

􀀀

include stdioh
define MaxNum      􀀀 List all primes upto MaxNum 􀀀
define NumIterations
 􀀀 Repeats procedure NumIterations times 􀀀
define TRUE
define FALSE 

void mainvoid

int IsPrimeMaxNum
int ikIteration 􀀀 Loop indexes 􀀀
int NumPrimes 􀀀 Number of primes found 􀀀
printfUsing Eratosthenes Sieve to find primes up to dn MaxNum
printfRepeating it d timesnNumIterations
for Iteration Iteration NumIterations Iteration

􀀀 Initialize all numbers to be prime 􀀀
for i i MaxNum i
IsPrimei TRUE
i
while ii MaxNum

if IsPrimei

􀀀 Mark all multiples of i to be nonprime 􀀀
k i i
while k MaxNum

IsPrimek FALSE
k k i
􀀀 of while k 􀀀
􀀀 of if IsPrime 􀀀
i i

􀀀 of WHILE ii 􀀀
NumPrimes  

for i i MaxNum i
􀀀 Count the number of primes 􀀀
if IsPrimei
NumPrimes NumPrimes
printfd primesnNumPrimes
􀀀 of for Iterations 􀀀
􀀀 The following can be added during debugging to list primes 􀀀
􀀀 for i  
i MaxNum i
if IsPrimei printfdni 􀀀

The result of running the program
Using Eratosthenes Sieve to find primes up to     
Repeating it
 times

     primes

     primes

     primes

     primes

     primes

     primes

     primes

     primes

     primes

     primes

􀀀 a Cannot compare systems oering dierent services
b 􀀀 Metric  response time
􀀀 Workload  Favourite programs  Word processor spreadsheet
c Metric  response time functionality Workload  A synthetic program
which tests the versions using various operating system commands
operating system services
d 􀀀 Metric  Response time reliability time between failures
􀀀 Workload  A synthetic program generating representative oppy
drive IO requests
e 􀀀 Metric  size of code structure of code execution time
􀀀 Workload  A representative set of programs in C and Pascal

􀀀 a
tCPU
􀀀
n
n
Xi􀀀
tCPU


􀀀
nIO
􀀀
n
n
Xi􀀀
nI􀀀O
􀀀

􀀀
s
xs
􀀀
n 􀀀
n
Xi􀀀

xsi xs

􀀀
n 􀀀 􀀀 n
Xi􀀀
x
si nx
s

􀀀 􀀀

􀀀
Similarly
s
xr


􀀀􀀀
b Normalize the variables to zero mean and unit standard deviation The
normalized values x􀀀s and x􀀀r are given by
x􀀀s
xs xs
sxs

xs 􀀀
􀀀
x􀀀r
xr xr
sxr

xr
􀀀􀀀
The normalized values are shown in the fourth and     fth columns of
Table b
The other steps are similar to example 􀀀
􀀀
Observation Variables Normalized Variables Principal Factors
No xs xr x􀀀s x􀀀r y y
􀀀 􀀀  􀀀􀀀
􀀀 􀀀􀀀 􀀀 􀀀
􀀀 􀀀
􀀀 
  􀀀  
􀀀 􀀀 􀀀
􀀀 􀀀  􀀀
Px 􀀀
Px 􀀀 􀀀
Mean
Standard
Deviation
􀀀 􀀀 􀀀 􀀀
The correlation between CPU time 
tCPU and number of IOs 
nIO
is The principal factors y and y are
􀀀 y
y 􀀀 p
p
p p
tCPU

nIO    
        
    
The     rst factor explains 􀀀􀀀
􀀀􀀀 or of total variation
There is no unique solution to this exercise Depending upon the choice of
outliers scaling technique or distance metric dierent results are possible
all of which could be considered correct One solution using no outliers range
normalization to 
􀀀 and Euclidean distance starts with the the normalized
values shown in the following
Program CPU time IOs
TKB 􀀀 􀀀
MAC
COBOL 􀀀
BASIC 􀀀
Pascal
EDT
SOS 􀀀
BASIC Pascal EDT COBOL SOS MAC and TKB join the dendro
gram at distances of 􀀀 􀀀 􀀀 and 􀀀􀀀 respectively