Software Design Methodology

Software Architecture Design 237

9.3.2 Evaluation and analysis of performance

To evaluate the performance of an architectural design in a particular scenario, the

information about the performances of each component and connector must be

available.

For example, consider the Keyword Frequency Vector extraction system

discussed in Chapter 8. To evaluate the performance of the architectural design in

the pipe-and-filter style on the following scenario, we must know the time required

by each processing component (i.e. the filters) to process each data element and the

time delay by the connectors (i.e. the pipes).

– Scenario 8: On average, a text file to be processed by the Keyword

Frequency Vector extraction system consists of N characters. The average

length of the words is about 6 characters. Among the words, there are about

40% small words that should be removed from the frequency calculation,

and the number of different keywords is only about 1% or less of the total

number of words.

Assume that the components and connector have the following performance

parameters

1

.

• The Input filter processes each character in 5 units of time.

• The Delete Small Words filter takes 20 units of time to decide whether the

word should be deleted from the stream or output to the stream.

• The Reduce Word to Original Form filter takes about 30 units of time to

process each word.

• The Sort Words Alphabetically component uses an insert sorting

algorithm to sort the words into alphabetical order. For each word it takes

about 3 log(n) units of time to insert it into the list, where n is the length

of the list of words.

• The Statistics of Frequency filter processes each word in 10 units of time.

• The Sort According to Frequency also uses an insert sorting algorithm and

takes 3log(n) time units to insert a keyword-frequency pair into the

keyword frequency vector of n elements.

• The Output component processes each element in 1 unit of time.

1

Notice that the parameters given in this example are for illustration of the

analysis and evaluation process only. It does not reflect the real performance of

a system.

238 Chapter 9. Scenario-Based Analysis and Evaluation

• There is a delay of about 10 units of time on all pipes in the system.

Figure 9.1 illustrates the performance parameters of the architectural design in

the pipe-and-filter style.

Figure 9.1 Performance parameters of pipe-and-filter architecture of KFV

Assume that the input stream of characters is always ready for the Input filter.

The Input filter will then produce words on its output stream at an average rate of 6

units of time per word. The first word produced by the Input filter will be 6 units

later after the start execution of the system. The whole stream of characters will be

processed by the Input filter in about N units of time, where N is the length of the

stream, i.e. the number of characters in the document. The earliest time that the

Input filter can finish processing of N characters is therefore, N units of time after it

started computation.

The Delete Small Words filter will start 10 units of time later after the Delete

filter output the first word. That is, it will start computation after about 16 units of

time after the whole system started execution. It will then take about 20 units to

process each word. Since the processing rate of the Delete filter is slower than the

upstream (i.e. the Input filter), the input to this filter can be considered always

ready. Consequently, it will produce output on its output stream at an average rate

of 28 units per word. The earliest time that it can produce a word to its output

stream is 36 units of time after the whole system started. It will take about 20 × N/6

Delete

small words

Reduce word to

original form

Output

Sort words

alphabetically

A stream of N

characters

A stream of

about N/6 words

A stream of N/10 words

without small words

A stream of N/10 words

in original forms

A stream of N/10 words

in alphabetical order

A stream of N/1000

word-frequency pairs

A keyword frequency

vector consists of

N/1000 elements

KFV in required

output format

Input

1u

1u

30u

20u

3lo

g

(

n

)

u

Statistics of

frequency

10u

Sort according

to frequency

3 log(n)u

10u

10u

10u 10u

10u

10u

10u