General Full Factorial Designs With Factors: Computer System Performance Anaylysis

You might also like

Download as ppt, pdf, or txt
Download as ppt, pdf, or txt
You are on page 1of 40

COMPUTER SYSTEM PERFORMANCE

ANAYLYSIS
GENERAL FULL FACTORIAL
DESIGNS WITH k FACTORS

1
Analysis of General Design
Informal Methods
 Observation Method
 Ranking Method
 Range Method

2
Total Number of Experiments – (Levels)Factors

(3)4 = 81

3
TYPES OF INTERACTIONS
TWO FACTOR 4C
2 6

THREE FACTOR 4C
3 4

FOUR FACTOR 4C
4 1

4
ANOVA
Table
For
Paging
Study

5
Component Sum of Percentage Degrees of Mean
squares of variation freedom square

Y 730.01 81

Y 681.21 1

y-y.... 48.80 100 80

6
MAIN EFFECTS

Component Sum of Percentage Degrees of Mean


squares of variation freedom square

(TOTAL) 45.80 93.85 8 5.7


Main Effects

A 1.30 2

D 6.10 2

P 12.30 2

M 26.20 2

7
FIRST ORDER INTERACTIONS
Component Sum of Percentage of Degrees of Mean square
squares variation freedom
TOTAL 2.40 4.91 24 0.1
(FIRST ORDER
INTERACTIONS)
AD 0.07 4
AP 0.02 4
AM 0.03 4
DP 0.15 4
DM 1.96 4
PM 0.14 4

8
SECOND ORDER INTERACTIONS
Component Sum of Percentage Degrees of Mean
squares of variation freedom square

TOTAL 0.48 0.98 32 0.015


(SECOND ORDER
INTERACTIONS)

ADP 0.05 8

ADM 0.13 8

APM 0.04 8

DPM 0.26 8

9
THIRD ORDER INTERACTIONS

Component Sum of Percentage Degrees of Mean


squares of variation freedom square

ADPM 0.07 0.14 16 0.004

10
Simplified Model to represent the paging process

yijkl=µ+αi+βj+γk+δl+ξjl
µ Grand Mean

αi Main Effect of A

βj Main Effect of D

γk Main Effect of P

δl Main Effect of M

ξjl Interaction Between


Factors D & M

11
ERRORS IN THE PAGING STUDY

12
Normal
Quatile-Quantile
Plot
Of Residuals

13
14
15
INFORMAL METHODS

Analysis Techniques presented so far produce


 Main effects
 Interactions
 Confidence Intervals

Goal : “Find the best combination of factor levels that produces


best performance”
Too much simpler techniques
 Results can be easily explained to decision makers with out
statistical details.

 Preliminary Analysis before ANOVA

16
OBSERVATION METHOD
RANKING METHOD
RANGE METHOD

17
OBSERVATION METHOD

If the Response variable – HB(Higher the Better) or


LB(Lower the Better) Metric
and Goal – Find the combination of factor levels that
produces best response

Simple look at Mean response Column

Experiments corresponding to Highest Response or Lowest


Response is found – Its levels give Desired Combination

Other Responses – Close to the Best Performance

Factor levels common to all such responses – DESIRED ANSWER

18
Example: Scheduler Design Study

Goal: Find Best Factor Levels that


Maximize the Throughput

3 Classes of jobs:
 Word Processing
 Interactive Data Processing
 Background Data Processing

16 experiments – 25-1 design


19
SYMBOL FACTOR LEVEL -1 LEVEL 1

A Preemption NO YES

B Time Slice SMALL LARGE

C Queue Assignment 1 QUEUE 2 QUEUES

D Requeueing 2 QUEUES 5 QUEUES

E Fairness OFF ON

20
21
SYMBOL FACTOR LEVEL -1 LEVEL 1

A Preemption NO YES

B Time Slice SMALL LARGE

C Queue Assignment 1 QUEUE 2 QUEUES

D Requeueing 2 QUEUES 5 QUEUES

E Fairness OFF ON

22
23
SYMBOL FACTOR LEVEL -1 LEVEL 1

A Preemption NO YES

B Time Slice SMALL LARGE

C Queue Assignment 1 QUEUE 2 QUEUES

D Requeueing 2 QUEUES 5 QUEUES

E Fairness OFF ON

24
25
SYMBOL FACTOR LEVEL -1 LEVEL 1

A Preemption NO YES

B Time Slice SMALL LARGE

C Queue Assignment 1 QUEUE 2 QUEUES

D Requeueing 2 QUEUES 5 QUEUES

E Fairness OFF ON

26
 RANKING METHOD 
Similar to Observation method

Experiments are written in order of increasing or decreasing


responses –
 Experiment with BEST response FIRST
 Experiment with WORST response LAST

Observe Factor columns – Find levels that consistently


produce good or bad results

More Information than Observation Method


27
28
29
SYMBOL FACTOR LEVEL -1 LEVEL 1

A Preemption NO YES

B Time Slice SMALL LARGE

C Queue Assignment 1 QUEUE 2 QUEUES

D Requeueing 2 QUEUES 5 QUEUES

E Fairness OFF ON

30
No A B C D E Tw Ti Tb
9 -1 -1 -1 1 -1 66.1
5 -1 -1 1 -1 -1 63.9
2 1 -1 -1 -1 -1 41.0
14 1 -1 1 1 -1 38.0
7 -1 1 1 -1 1 36.3
3 -1 1 -1 -1 -1 36.0
15 -1 1 1 1 -1 35.0
11 -1 1 -1 1 1 34.6
13 -1 -1 1 1 1 26.0
1 -1 -1 -1 -1 1 25.0
8 1 1 1 -1 -1 23.0
12 1 1 -1 1 -1 23.0
16 1 1 1 1 1 22.0
4 1 1 -1 -1 1 15.7
6 1 -1 1 -1 1 13.2
10 1 -1 -1 1 1 9.2
31
SYMBOL FACTOR LEVEL -1 LEVEL 1

A Preemption NO YES

B Time Slice SMALL LARGE

C Queue Assignment 1 QUEUE 2 QUEUES

D Requeueing 2 QUEUES 5 QUEUES

E Fairness OFF ON

32
No A B C D E Tw Ti Tb
3 -1 1 -1 -1 -1 21.0
7 -1 1 1 -1 1 20.2
15 -1 1 1 1 -1 17.2
11 -1 1 -1 1 1 15.7
1 -1 -1 -1 -1 1 15.2
13 -1 -1 1 1 1 12.0
4 1 1 -1 -1 1 8.6
10 1 -1 -1 1 1 8.4
5 -1 -1 1 -1 -1 7.5
6 1 -1 1 -1 1 7.5
9 -1 -1 -1 1 -1 6.4
2 1 -1 -1 -1 -1 3.0
8 1 1 1 -1 -1 3.0
12 1 1 -1 -1 1 3.0
14 1 -1 1 1 -1 2.0
16 1 1 1 1 1 2.0
33
SYMBOL FACTOR LEVEL -1 LEVEL 1

A Preemption NO YES

B Time Slice SMALL LARGE

C Queue Assignment 1 QUEUE 2 QUEUES

D Requeueing 2 QUEUES 5 QUEUES

E Fairness OFF ON

34
RANGE METHOD

Importance of a factor in a model -> Find percentage of variation


explained by that factor

Informal alternative:
Find the average response
Find the difference between max & min -> RANGE

Factor with large range -> IMPORTANT

Example: Data of paging study

Calculate averages corresponding to three levels of each


of the four factors

35
LEVEL
FACTOR
1 2 3

Replacement Algorithm 2056 2986 3781

Deck Arrangement 1584 2913 4326

Problem Program 592 2047 6185

Memory size 305 2006 6512

36
55523 27 = 2056
37
8233 27 = 305
38
LEVEL Range of
FACTOR averages
1 2 3

Replacement Algorithm 2056 2986 3781 1725


Deck Arrangement 1584 2913 4326 2742
Problem Program 592 2047 6185 5593
Memory size 305 2006 6512 6207

****MEMORY SIZE – Most influential factor


***PROBLEM PROGRAM
**DECK ARRANGEMENT
*ALGORITHM

39
The End
and
Thank You
40

You might also like