A Rationale-based Model for Architecture Design Reasoning

More documents

Recommendations

Info

10.1. Background and the dependencies between architecture elements and architecture rationale in an AREL model. 10.1.1 Related work Fenton and Neil [40] believe that management decision support tools must be able to handle causality, uncertainty and combining different (often subjective) evidence. Hence they suggest to use a solution based on BBNs. BBNs have been used in many applications (see [83] for a recent survey) including a causal model to detect user interactions with user interface in safety-critical systems which are prone to human errors [45]. A design reasoning approach has been proposed by Zhang and Jarzabek to use BBN to reason about architecture design decisions on quality attributes [177]. The purpose of this work is to use BBN as an aid to evaluate the architecture design for decision making. BBNs [116, 73] are graphical models for probabilistic reasoning, which are now widely accepted in the AI community as practical and intuitively appealing representations for reasoning under uncertainty. Given the BBN representation of AREL, we show how BBN reasoning algorithms can be used to reason about change impacts. More specifically, architects can use BBN to undertake what-if analysis, to predict and diagnose impacts given complex combinations of requirement and design changes. This approach has the following advantages: • We represent the AREL model as a BBN. The BBN graphical structure models and quantifies the causal relationship between architecture design decisions and architecture elements. • Based on the AREL causal relationship in a BBN model, we employ three different probability-based reasoning methods to carry out change impact analysis: – Predictive Reasoning - predict what design elements and how likely they might be affected if one or more architecture elements are to change (e.g. requirements). – Diagnostic Reasoning - if one or more architecture elements (e.g. design object) are to change, diagnose what might be the causes of such change and the extent of their influence. – Combined Reasoning - by combining the use of predictive reasoning and diagnostic reasoning, reason about the ripple effect of the likely changes to the system through diagnosing the causes and predicting the effects. 167
10.1. Background 10.1.2 Introduction to Bayesian Belief Networks Bayesian Belief Networks (BBNs) are a well established method in the Artificial Intelligence community for reasoning under uncertainty, using (i) a graphical structure to represent causal relationships and (ii) probability calculus to quantify these relationships and update beliefs given new information. A BBN is a directed graph with edges connecting nodes with no cycles (i.e. a directed acyclic graph). Its nodes represent random variables and its edges represent direct dependencies between variables. In theory, variables can be discrete or continuous, but in this work we construct models with discrete variables only. Nodes representing discrete variables have two or more discrete states, representing the possible values the variable may take. For instance, a discrete Boolean node may represent whether or not a patient has lung cancer with the two states True and False. Nodes without parents are called root nodes and have an associated prior probability distribution. Each node with parents has a conditional probability table (CPT), which, for each combination of the values of the parents, gives the conditional probability of its taking these values. Thus the CPT can be considered to quantify the strength of the causal relationships. Let us illustrate these basic elements of a BBN – the nodes, arcs and prior and conditional probabilities – with a very simple example from the medical domain [83]. Suppose that a physician wants to reason about the chance that a patient presenting with a cough has lung cancer. The first relevant causal factor needing to be established would be whether the patient is a light, heavy or non-smoker. A possible diagnostic tool would be to take an Xray of the lungs. Figure 10.1(a) shows a BBN model for this example. The example illustrates that a heavy smoker has a higher probability of contracting lung cancer than a non-smoker. If a person has lung cancer, than there is a high probability that the person would cough, and there is a high probability that the lung cancer would be detected by the X-ray. Once a BBN model has been constructed, posterior probabilities – often referred to as beliefs – can be computed. Figure 10.1(b) shows the physician’s prior beliefs before any observations or evidence are taken into account. The marginal probability distribution is represented by the bars in the diagram. Users can set the values of any combination of nodes in the network and this newly inserted evidence propagates through the network through computing their marginal probabilities, producing a new probability distribution over all the variables in the network. An explanation on the computation of marginal probabilities can be found in [83, 73, 117]. There are a number of efficient exact and approximate inference algorithms for performing this probabilistic updating [117], providing a powerful combination of predictive, diagnostic and explanatory reasoning. For example, 168
Page 1 and 2:
A Rationale-based Model for Archite
Page 3 and 4:
maintenance. These techniques can h
Page 5 and 6:
Declaration This thesis contains no
Page 7 and 8:
2.2.6 DoD Architecture Framework .
Page 9 and 10:
5.4.6 Risk assessment in architectu
Page 11 and 12:
8.2 The architecture rationalisatio
Page 13 and 14:
12.2 Contributions . . . . . . . .
Page 15 and 16:
6.10 ≪AR≫ and ≪AAR≫ Stereot
Page 17 and 18:
11.13A Process to Synchronise Chang
Page 19 and 20:
5.13 Design Rationale Helps Evaluat
Page 21 and 22:
that architecture design is highly
Page 23 and 24:
1.2. Research motivations and resea
Page 25 and 26:
1.2. Research motivations and resea
Page 27 and 28:
1.4. Structure of the thesis the ch
Page 29 and 30:
Part I Problem Analysis 10
Page 31 and 32:
2.1. Industry practice of design ra
Page 33 and 34:
2.2. Architecture frameworks and de
Page 35 and 36:
Page 37 and 38:
Page 39 and 40:
Page 41 and 42:
Page 43 and 44:
Chapter 3 Related work in design ra
Page 45 and 46:
3.2. Why do we need design rational
Page 47 and 48:
3.2. Why do we need design rational
Page 49 and 50:
3.3. Existing methods for capturing
Page 51 and 52:
Page 53 and 54:
Page 55 and 56:
Page 57 and 58:
3.4. Other related studies 3.3.7 Qu
Page 59 and 60:
3.4. Other related studies evolutio
Page 61 and 62:
3.5. How well design rationale meth
Page 63 and 64:
3.5. How well design rationale meth
Page 65 and 66:
Chapter 4 Research methodology and
Page 67 and 68:
4.1. Software engineering research
Page 69 and 70:
4.2. The chosen research and valida
Page 71 and 72:
Part II Architecture Design Rationa
Page 73 and 74:
on a regular basis. Their inputs ha
Page 75 and 76:
5.1. Architecture rationale in the
Page 77 and 78:
5.2. Survey methodology data collec
Page 79 and 80:
5.3. Survey findings experience). I
Page 81 and 82:
5.3. Survey findings Table 5.1: Fre
Page 83 and 84:
5.3. Survey findings Business Goals
Page 85 and 86:
5.3. Survey findings 5.3.5 Document
Page 87 and 88:
5.3. Survey findings Table 5.8 summ
Page 89 and 90:
5.3. Survey findings We used Spearm
Page 91 and 92:
5.3. Survey findings previous desig
Page 93 and 94:
5.4. Discussion of findings Since t
Page 95 and 96:
5.4. Discussion of findings or cons
Page 97 and 98:
5.5. Limitations and use design rat
Page 99 and 100:
Part III The Representation and App
Page 101 and 102:
6.1. A conceptual model for design
Page 103 and 104:
6.2. Architecture Rationale and Ele
Page 105 and 106:
6.2. Architecture Rationale and Ele
Page 107 and 108:
6.3. Architecture elements Figure 6
Page 109 and 110:
6.4. Architecture rationale Figure
Page 111 and 112:
6.4. Architecture rationale • ris
Page 113 and 114:
6.4. Architecture rationale depicts
Page 115 and 116:
6.5. The extended AREL To prevent c
Page 117 and 118:
6.6. A UML representation of AREL a
Page 119 and 120:
6.6. A UML representation of AREL a
Page 121 and 122:
6.7. AREL usability AE by ≪AEsupe
Page 123 and 124:
6.8. Summary The AREL model is exte
Page 125 and 126:
7.1. The EFT system reasoning, a co
Page 127 and 128:
7.1. The EFT system The processing
Page 129 and 130:
7.1. The EFT system not chosen beca
Page 131 and 132:
7.1. The EFT system Figure 7.5: Int
Page 133 and 134:
7.1. The EFT system sequence checki
Page 135 and 136: 7.1. The EFT system Figure 7.7: Dec
Page 137 and 138: 7.1. The EFT system Therefore, in a
Page 139 and 140: 7.1. The EFT system Figure 7.10: De
Page 141 and 142: 7.2. An empirical study to validate
Page 151 and 152: 7.3. Summary The case study is spec
Page 153 and 154: 8.1. Background Based on these assu
Page 155 and 156: 8.2. The architecture rationalisati
Page 163 and 164: 8.3. Other applications of ARM As d
Page 165 and 166: 8.3. Other applications of ARM Figu
Page 167 and 168: 8.3. Other applications of ARM area
Page 169 and 170: Chapter 9 Architecture rationale an
Page 171 and 172: 9.1. Background 9.1 Background In t
Page 173 and 174: 9.2. Traceability of architecture r
Page 175 and 176: 9.4. AREL and eAREL traceability ap
Page 185: Chapter 10 Architecture decision de
Page 189 and 190: 10.2. Building a BBN to represent a
Page 197 and 198: 10.3. Reasoning about change impact
Page 209 and 210: 10.5. Summary design object may cau
Page 211 and 212: 11.1. Capturing architecture design
Page 213 and 214: 11.1. Capturing architecture design
Page 215 and 216: 11.2. Checking AREL models Figure 1
Page 217 and 218: 11.3. Tracing AREL models Figure 11
Page 219 and 220: 11.4. Analysing AREL with BBN • S
Page 221 and 222: 11.5. Limitations Figure 11.13: A P
Page 223 and 224: Chapter 12 Conclusions 12.1 Summary
Page 225 and 226: 12.2. Contributions reasoning metho
Page 227 and 228: 12.2. Contributions implementations
Page 229 and 230: 12.3. Future work architecture deci
Page 231 and 232: 12.3. Future work to examine the co
Page 233 and 234: BIBLIOGRAPHY [10] B. W. Boehm, Soft
Page 235 and 236: BIBLIOGRAPHY [37] A. Eden and R. Ka
Page 237 and 238:
BIBLIOGRAPHY [62] J. D. Herbsleb an
Page 239 and 240:
BIBLIOGRAPHY [89] N. Lassing, D. Ri
Page 241 and 242:
BIBLIOGRAPHY [115] D. L. Parnas,
Page 243 and 244:
BIBLIOGRAPHY [142] Z. Simsek and J.
Page 245 and 246:
BIBLIOGRAPHY [167] W. F. Tichy, N.
Page 247 and 248:
Appendix A AREL Tool User Manual 1
Page 249 and 250:
4.1 AREL Check Model When the Check
Page 251 and 252:
4.2 AREL Model Tracing AREL model t
Page 253 and 254:
Figure 9 - result of downward trace
Page 255 and 256:
To initiate the graph’s creation,
Page 257 and 258:
onto the diagram 14 Assign values t
Page 259 and 260:
Useful Definitions The following ar
Page 261 and 262:
) Some projects only c) Not at all
Page 263 and 264:
19. I revisit architecture design d
Page 265 and 266:
Complaint Procedure If you have any
Page 267 and 268:
Privacy Protection Only investigato
Page 269 and 270:
B. Exploring design reasoning. 1. W
Page 271 and 272:
3. Why does the system use asynchro
Page 273 and 274:
C. General Comments on AREL Modelli
Page 275:
IEEE Computer Society Press. A. Tan
show all

A Rationale-based Model for Architecture Design Reasoning

Create successful ePaper yourself

Delete template?

Save as template?