A Formal Approach For Change Impact Analysis of Long Term Composed Services Using Probabilistic Cellular Automata

JKSUCI 206 No.
of Pages 19
20 November 2015
Journal of King Saud University – Computer and Information Sciences (2015) xxx, xxx–xxx
1
King Saud University

Journal of King Saud University –
Computer and Information Sciences
www.ksu.edu.sa
www.sciencedirect.com
3 A formal approach for change impact analysis

4 of long term composed services using Probabilistic
5 Cellular Automata
6 M. Thirumaran, M. Jannani *, N. Sivakumar
7 Pondicherry Engineering College, Puducherry, India
8 Received 11 January 2014; revised 26 June 2014; accepted 23 October 2014

9
11 KEYWORDS Abstract Incorporating changes into the logics of composed services dynamically and successfully
12
13 Service oriented architecture; is a challenge for sustaining a business’ image and profit in the society, especially when the change is
14 Web services; expected to be made immediately at low cost. In this paper, we address this challenge by proposing
15 Change management; a change impact analysis framework for long term composed services (LCS) which: (i) enables the
16 Change impact analysis; business people to implement the changes by themselves through their analysts, (ii) reduces cost and
17 Finite State Machine; time by eliminating the dependence on IT developers once the application services are developed
18 Probabilistic Cellular and delivered, (iii) ensures effective incorporation of the changes made by using standard method-
19 Automata ologies for evaluation – finite state automaton for verifying the runtime compatibilities and change
evaluation and probabilistic cellular automaton for impact analysis and prediction. Through the
evaluated probability measures and effective incident matching, the knowledge gained by the ana-
lyst over his service logics and the efficiency of incorporating changes are increased.
20 Ó 2015 Production and hosting by Elsevier B.V. on behalf of King Saud University. This is an open access
article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
21 1. Introduction the partnership among the services in composition and the 28

business objective to be attained are for a long run. LCSs facil- 29
22 Service Oriented Computing (SOC) is emerging as a new para- itate dynamic modification of the composition structure by 30
23 digm and is accruing the outsourcing of the required function- choosing the best services and dissolving the services whose 31
24 alities from third party web based providers especially through execution is no longer needed. They thus enable dynamic selec- 32
25 service composition. Composite services are built from aggre- tion of the partners and aid the end users and consumers to be 33
26 gates of other autonomous services which collectively provide highly benefited from the open competition among the busi- 34
27 a value added service. In a long term composed service (LCS), nesses. As the end users are always interested in latest tech- 35
niques and technologies, the LCSs are more prone to 36
* Corresponding author. changes, as time passes. Changes to an LCS can be of two 37
Peer review under responsibility of King Saud University. types: top down changes or bottom up changes (Liu et al., 38
2011). In the former, the changes are initiated by the owners 39
of the LCS and in the latter they are originated from the out- 40
sourced service providers. With the increase in the emanation 41
Production and hosting by Elsevier of wide range of business competitors and the demands of 42
http://dx.doi.org/10.1016/j.jksuci.2014.10.009
1319-1578 Ó 2015 Production and hosting by Elsevier B.V. on behalf of King Saud University.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
Please cite this article in press as: Thirumaran, M. et al., A formal approach for change impact analysis of long term composed services using Probabilistic Cellular
Automata. Journal of King Saud University – Computer and Information Sciences (2015), http://dx.doi.org/10.1016/j.jksuci.2014.10.009
JKSUCI 206 No. of Pages 19
20 November 2015
2 M. Thirumaran et al.
43 the end users, effective change management in LCSs has matched with the pre occurred, similar change request patterns 105
44 become vital. The dependence on the IT developers for the followed by the behavioral analysis which is performed using 106
45 incorporation of these changes claims the investment of a con- Probabilistic Cellular Automata (PCA). PCA makes use of 107
46 siderable amount of time and cost which poses a significant the knowledge extracted from these patterns and indicates 108
47 threat to the business’ income. the risk, degree of automation, accuracy and degree of incident 109
48 So there is a need for a framework which would enable the matching involved which would aid the analyst in making the 110
49 business people to make changes to the LCS through their ana- changes confidently and maximize the market attraction of the 111
50 lysts without the aid of the IT developers. Once a change LCS. 112
51 occurs, the framework must enable the LCS to adapt itself
52 quickly and automatically in order to satisfy the business need. 2. Change impact analysis of LCS using Probabilistic Cellular 113
53 However such changes should be incorporated systematically Automata 114
54 without any issues (Maamar et al., 2008) and hence change
55 impact analysis which involves analyzing the impact which In this section, we present our change impact analysis frame- 115
56 the incorporated change would have on the LCS, is essential work for managing changes in LCS using Probabilistic Cellu- 116
57 and would increase the accuracy of the changes made. Though lar Automata. We first illustrate the working of the change 117
58 there are a lot of research works focusing on automatic com- impact analysis framework shown in Fig. 1 by means of the 118
59 position and integration of the services, change impact analysis sequence of operations and activities involved in the change 119
60 in web service change management using formal approaches impact analysis and their dependencies depicted in Fig. 2. 120
61 has not been lime lighted much. The existing frameworks do We then describe how the framework reduces the cost and time 121
62 not provide maximum runtime support and degree of automa- involved in effectively incorporating the changes with the help 122
63 tion and do not enable changes at the analyst level (Apostolou of a motivating example of a sample web service composition. 123
64 et al., 2010). They do not perform change impact analysis and The impact of a change, the construction of LCS schema, FSM 124
65 incident matching which are very much essential for a timely and the associated PCA for the example are demonstrated 125
66 and cost effective change management (Rovegard et al., manually. LCSs have attracted a lot of attention since they 126
67 2008; Setzer et al., 2010; Chua and Aslam Hossain, 2012). provide a powerful tool to offer value-added and customized 127
68 There are frameworks for handling top down and for handling services. These vividly contribute toward the need to standard- 128
69 bottom up changes but not for both. There are no standard ize and fine tune the change management process for LCSs and 129
70 procedures and methodologies adopted in the existing works to follow structured approaches to make error free changes in 130
71 and even if the changes are made, there is no means to find the functionality of the services without compromising the 131
72 the exact reason for the change being unmanageable in the cur- quality of the business process. 132
73 rent static scenario. In the current situation it is very difficult
74 to incorporate the changes perfectly into the logics of the 2.1. Overview of change impact analysis approach and 133
75 LCS. All these together have motivated our research question, framework 134
76 ‘‘How can the change management process in long term com-
77 posed services be enhanced such that the changes can be made
78 by the analysts in a timely and cost effective manner and with The importance of the change impact analysis approach can be 135
79 reduced risk?”. adumbrated with the help of the following example. Consider 136
80 In this paper, we have addressed this issue through the pro- a business analyst working on services in long term composi- 137
81 posed change impact analysis framework which adopts formal tion. In case of a situation where a change has to be made to 138
82 methods and standard procedures for analyzing the impact of the service logic immediately and at low cost, the analyst is 139
83 the changes through effective runtime support, change evalua- put in a situation where he cannot wait for the development 140
84 tion, constraint evaluation and QoS factor evaluation. When a team and has to make the changes by himself. So, there are 141
85 request for making a change is received, the framework verifies high chances for him to make a bug introducing change and 142
86 if there are no runtime issues in the business logic extracted by inject incorrect statements into the logic of the LCS which 143
87 property pre evaluation using Finite State Machine (FSM) and might end with the changed LCS exhibiting an undesired 144
88 only then it allows the analyst to make the changes over the behavior. Even if the changes are made as careful as possible, 145
89 logic. After the changes are made, the change evaluation is without analyzing the impact that the changes would have on 146
90 done which verifies if the change made can be committed safely the logic, it is not possible to assure a risk-free and accurate 147
91 or not. This involves the checking over the deviation of the incorporation of the changes which in turn serves as a serious 148
92 functionality of the logic after change through change factor threat to meet the business outcome. This problem is addressed 149
93 evaluation, over the workflow of the logic through constraint by the change impact analysis framework depicted in Fig. 1. 150
94 factor evaluation and over the quality of the modified LCS Initially the received request for change is sent to the 151
95 using QoS factor evaluation for differentiating from competing change request manager to identify the part of the business 152
96 service providers. All these evaluations are performed using process in which the change has to be made. The change 153
97 FSM which provides the formalism required for verifying the request consists of the command to be executed i.e. the 154
98 changes made. The impact that the changes would have on DML operations to be performed like an addition or a substi- 155
99 the LCS is indicated through impact analysis and incident tution etc., the resource to which the change is to be done 156
100 matching. The evaluated factors are compared with the thresh- expressed in terms of the process name and the condition to 157
101 old estimations and impact value estimations from the previ- be followed i.e. performing the mentioned steps on the satisfac- 158
102 ous incidents and Cellular Automata (CA) is used to validate tion of certain condition mentioned. Then, the source code of 159
103 the evaluation and to generate the change factor, constraint the composed service which can be in any language, to which 160
104 factor and QoS factor patterns. The current changes are the change is to be made, is extracted by analyzing the change 161
20 November 2015
Change impact analysis of long term composed services 3
Figure 1 Change impact analysis framework.
162 request as shown in step 1 of Fig. 2. The domain analysis is tion set, composition type and the process owner (policy 177
163 performed to identify the business domain and the context dependencies are checked) are analyzed and checked. The ser- 178
164 analysis is performed to identify the business environment of vice registry maintains information about the services includ- 179
165 the composed service. The business logic analysis involves ing the service id, service profile and the service location. 180
166 the analysis of the project domain, business process and the The business logic source maintains the business logic set, busi- 181
167 business logic (rule, function). ness logic schema and the corresponding Finite State Machine, 182
168 This is followed by the dependency analysis. The entities state transition table and the change measure table. 183
169 are checked for any input dependency where two or more enti- After the dependency analysis, the rules, functions, param- 184
170 ties depend on the same input, output dependency where enti- eters and dependency are identified and are placed in the busi- 185
171 ties are expected to produce the same output, call sequence ness logic set as shown in step 2 of Fig. 2. Rules, their syntax 186
172 dependency, policy dependency where the entities are expected and their description are identified. Functions are expressed in 187
173 to abide by the same policy, mapping dependency where enti- terms of primitive computable functions. Then the LCS 188
174 ties depend on the same data set for instance and existence schema before change is identified and generated by the 189
175 dependency where the existence of an entity is meaningless in schema generator as shown in step 3 of Fig. 2. Business logic 190
176 the absence of another particular entity. The service composi- is a group of the logic entities which includes rules, functions, 191
20 November 2015
Change Request
Source Code
[Command] 1 @WebService(serviceName= 2 Business Logic Set
”Singapore_Airlines”) Service Rule Param Dependency
<Resource>
ResultSet set Set Set Set
(Condion) rs1=stmt.executeQuery(“Sel {S1,S2} {R1,R2} {P1,P2} {D1,D2}
ect * from airline where
3
Property Before Aer FSM Construcon (before Business Logic Schema

Traceability True 5 Change) 4
<LCS Schema [Airline][Hotel][Cab]>
Accessibility True <cservice>
Computability True <service name=Singapore_Airlines”>
Dependency True <rule id=”r11” name=”getcust_details”>
<input>
6 <parametername=”name”
ptype=”String”/>
Change Factor Aer
Change
Order of Execuon 80%
QoS Factor Before Aer Property Before Aer
Correctness 89%
Change Change Traceability True True
Similarity Measure 56% 7 8
Execuon Time 223 ms 220 ms
Business Policy Enforcement 77% Accessibility True True
Parallel True Availability 98% 99% Computability True True
Branch True Cost $ 669 $ 660
Dependency True True
Iteration True Response Time 340 ms 338 ms
Synchronizaon True
9
Threshold name Impact Impact Threshold Decidabil- Name Type Impact

priority value value 10 ity Factor Value
Threshold Very High, TVH 5 40% 100 IOOE Order of Change 3
11 Threshold High, TH 4 45% 90 Execuon Factor
Threshold Medium, TM 3 50% 88 ISM Similarity Change 2
Threshold Low, TL 2 55% 34 Measure Factor
…
12
13
CR CR LCS
Probability Esmated probability
Timestamp Domain Condition Status esmaon from measures
ID Type Set 14
2012/3/12 Travel Policy not
{Airline,
Change
incidents
12 6:56:50 PM Deletion Hotel, Degree of automaon 0.723
Agency violated Committed
(UTC+05:30) Cab}
Risk 0.34
Incident Matching 0.89
Accuracy 0.666
Figure 2 Workflow diagram of change impact analysis framework.
192 parameters and dependency. Every operation in a web service the complete schema. After the schema validation is done, 197
193 is rule based and is reflected by the schema. Every rule is con- the corresponding FSM which provides better understandabil- 198
194 verted into schema and this reduces the time taken in the gen- ity and is language neutral is generated as shown in step 4 of 199
195 eration of schema at runtime. The schema reflects the part of Fig. 2. FSM being a conceptual and machine processable 200
196 the logic under change only and hence is validated against model provides enhanced focus on the part of the business 201
20 November 2015
P
202 process where the change has to be made by reflecting the in M is given by, d(qn, n) ? Fn, where qn 2 Qn and Fn # Qn 259
203 schema. Any change made to the FSM is reflected back in is set of final states. This verification checks if there is a rule 260
204 the schema. While FSM consists of nodes, the schema consists initiating the execution (start state) and for completing the exe- 261
205 of Meta data which are very useful when fetched during back- cution (end state). The traceability property is verified in order 262
206 tracking. In order to fetch the Meta data, the information to make sure that the business logic entity (service, rule, func- 263
207 about the node is needed which is provided by the FSM. This tion or parameter) in the change request is traceable in the con- 264
208 provides better understandability to the business analyst and structed FSM as the FSM reflects only the part of the logic 265
209 the analyst is facilitated to make the change which is then under change. The services S1, S2, S3 . . . Sn are said to be 266
210 reflected back at the source code. For an analyst with little dependent P on each other P when the
T transitionP function is as 267
211 knowledge about the code, this facilitation would grant 9n
P 4(S n , n) ? d(Sn+e , n+e ) 4(S n+e , n+e) ? d(Sn+d, 268
212 immense comfort. But still, the changes must be carefully eval- n+d), where e > 0 and d > 0. The accessibility property is 269
213 uated, verified and only then reflected in the source code to checked in order to verify that the business logic entities have 270
214 avoid critical latent issues which is assured by the proposed the access permission to incorporate the change. 271
215 framework. This is guaranteed by the framework by perform-
216 ing the Property evaluation, change factor evaluation, con- 2.3. Change management 272
217 straint factor evaluation, QoS factor evaluation and change
218 impact analysis. After the property evaluation before change is done, the ana- 273
lyst is allowed to perform functional changes to the logic. 274
219 2.2. Runtime management Then, the corresponding FSM after change is constructed 275
and the change evaluation is done. The change evaluator per- 276
220 The property evaluator performs the process of evaluating the forms the process of change factor evaluation, constraint fac- 277
221 identified vital properties. The framework manages the tor evaluation and QoS factor evaluation. This assures that 278
222 changes dynamically without any exceptions and makes sure the changes made to the business logic are in accordance with 279
223 that the changes are made as expected, with the help of this the change request, meeting the expectation of the analyst and 280
224 property evaluator. When a part of the logic is extracted from that the changes are injected correctly into the logic. 281
225 an entire business process of the LCS, it tends to have compat-
226 ibility issues to implement a change like its dependence on a 2.3.1. Change factor evaluator 282
227 rule, function, parameter of change. The extracted logic might Change factors are envisaged as the touchstones for the evalu- 283
228 not be complete with all required rules to implement the ation of the changes made. Any change in the business logic is 284
229 required change and some required rules or functions might influenced by the change factors either independently or in a 285
230 be missing. The constructed FSM might reflect the rules, func- group and they aid in making the changes effectively. The vital 286
231 tions or even services which do not have the access permission factors for change factor evaluation are listed in Table 1. Order 287
232 to make a change or the rules might make a call to the rules of P DRid P DFid
of execution is given by OOE ¼ R
[ F
100, 288
233 other services. All these issues need to be checked to set the
where Rid, Fid are the unique ids which establish the order of 289
234 manageability factor which indicates whether the extracted
the rule and the function within the logic L. Similarity measure 290
235 logic and the constructed FSM is complete reflecting the exact P DIP =DOP
236 code to be changed and hence fit for a change process as shown is estimated by SM ¼ N
100, where DIP and DOP, 291
237 in step 5 of Fig. 2. refer to the number of states in which the inputs and outputs 292
238 Property evaluation is done all the time during the schema have varied respectively and N is the total number of states 293
239 extraction, during the change process and even after the generated for L in FSM, M (Rovegard et al., 2008). Business 294
P FP
240 change is committed or discarded. The properties including Policy Enforcement is given by BPE ¼ 100, where 295
P
241 computability, accessibility, traceability and dependency are FP is the states in which the business policies failed and P 296
242 evaluated over the FSM and the property evaluation tag is are the total states related to business policies in M. Correct- 297
243 updated in the business logic schema. The evaluation of the Ph i
ness is given by 1 CTds 100, where Cd is the number
244 computability property involves the check over the reachabil- 298
245 ity and completeness of the logic reflected by the FSM. Let of dangling states and Ts is the total number of states in M. 299
246 S1, S2, S3 . . . Sn be the services in composition. Let M be the The evaluation of these change factors indicates the deviations 300
respective Finite caused in the functionality of the business logic due to the
P State Machine associated with it. M is defined
247 301
as MP= {Q, , d, qo, F}. The P transition function is given change as shown in step 6 of Fig. 2.
P as, d
248 302
249 (Q1, 1) ? {Q P2 }, d(Q 2 , 2) ? {Q 3}, . . . d(Q n1 , n1) ?
250 {Qn}, where n is the set of input alphabetsPfor the Sn for 2.3.2. Constraint factor evaluator 303
251 any n, such that Q1 consumes P the input set 1, goes to Q2 Constraint factors play a vital role in the evaluation of the 304
252 and finally Qn1 consumes n1 and reaches Qn. If this is sat- changes made by evaluating the changes with respect to the 305
253 isfied, reachability property is observed to be true. This verifi- invocation order and the work flow order which is significantly 306
254 cation makes sure that every node i.e. every rule or function important in case of composed services. Synchronization, 307
255 represented in the FSM is reachable from every other rule or Sequence, Branch, Parallel and Iteration are the constraint fac- 308
256 function. This identifies the lack of call to any rule or function tors considered for the functional changes made to the busi- 309
257 of the logic under change. Completeness is associated with the ness logic. Let S1, S2, S3 . . . Sn be the set of n services in 310
258 single service wherein the transition function for any service Sn composition comprising of rule sets such as S1 = {r1, r2, 311
20 November 2015
in a service reaches the same state qn using the logic L then iter- 332
Table 1 Evaluation factors and their purpose.
ation is said to be observed. The factors are evaluated and 333
Evaluation Component Factors Purpose updated as shown in step 6 of Fig. 2. 334
phase evaluated
Runtime Property Computability In order to provide 2.3.3. QoS factor evaluator 335
management evaluator Traceability runtime support by QoS factors influence the changes made to the composite ser- 336
Accessibility checking if the FSM
vice in the idea that any service in composition with devastat- 337
Dependency reflects the exact
logic under change
ing variations in its non functional behavior after change is not 338
and is free from preferable as it might result in the failure to meet the business 339
compatibility issues competitors (Xiong et al., 2009; Wang et al., 2009). So when a 340
before and after change has been made to the logic, the quality metrics pave the 341
change way for making the behavioral analysis and committing the 342
Change Change Order of In order to evaluate change. Response time, execution time, cost and availability 343
management factor execution the changes made to are the vital QoS factors evaluated. Web service availability 344
evaluator Correctness the functionality of is the ratio of the expected value of the uptime of a service 345
Similarity the business logic to the aggregate of the expected values of up and down time. 346
measure
Response time is the measure of how long it takes for a service 347
Business
policy
to respond from the time of invocation of the service. Execu- 348
enforcement tion time is the measure of the time taken for the complete exe- 349
Constraint Sequence To evaluate the cution of a rule or service in composition. Cost is measured 350
factor Parallel changes made with from the estimation of the execution time of the rule or service 351
evaluator Branch respect to the in composition. The QoS factors are evaluated and updated as 352
Iteration invocation and shown in step 7 of Fig. 2. After these evaluations are done, the 353
work flow order property evaluation is done again as shown in step 8 of Fig. 2 354
QoS factor Response time In order to verify and the change evaluation values are updated as reference 355
evaluator Execution the impact that the points in the business logic schema before the change is com- 356
time change has over the
mitted. Based on these values, the impact analysis is done. 357
Cost quality of the
Availability business process
Temporal analysis is performed which involves the behavior 358
Change Behavior Probability of To deduce the analysis and growth rate analysis which focuses on the profit 359
impact analyzer risk measures for or loss gained from the composition. 360
analysis Probability of decision making and
accuracy analyze the behavior 3. Change impact analysis 361
Probability of of the composed
incident service after change
matching using PCA The change, constraint and the QoS factors together are called 362
Probability of the decidability factors which determine whether the change 363
degree of made can be committed or not through the estimation of 364

automation decidability. For every change request once the decidability 365
factors are measured, they are considered for impact analysis. 366
The amount of impact (impact factor) of each decidability fac- 367
tor on the logics is determined first. The
P generalized formula 368
312 r3 . . . rp}, S2 = {rp, rp+1, rp+2 . . . rq}, . . . Sn = {rq+1, rq+2, rq+3 for determining impact factor is: If ¼ ðDfv Iv Þ=NDf where, 369
313 . . . rr}. These services are said to be synchronized with each If – impact factor, Dfv – decidability factor value, Iv – impact 370
314 other (S1 [ S2 [ S3 . . . Sn) if and only if S1 [ S2 = ;, value, NDf – total number of decidability factors. The decid- 371
315 S2 [ S3 = ; . . . Sn1 [ Sn = ;. The sequence is said to be ability factor values (Dfv) are taken from the measures done 372
316 observed in LCS if and only if, 9n d in the change, constraint and the QoS evaluators. The impact 373
317

ðSn ; DSn Þ ¼> Snþe \ d ðSnþe ; DSnþe Þ ¼> Snþd where d values (Iv) for each decidability factor is provided in Table 2. 374
318 ðSn ; DSn Þ ¼> Snþe indicates the transition of Sn to Sn+e taking Impact value is the priority of impact that a particular decid- 375
319 DSn input in multiple steps, e > 0 and d > 0, DSn represents ability factor has on the business logic which is estimated from 376
320 the output of Sn. Let S1, S2, S3 . . . Sn be n services possessing the behavior of the factor in the previous similar incidents. The 377
321 the states Q1, Q2, Q3 . . . Qn. The branch is said to be observed decidability factors are rated from 1 to 5. For example con- 378
322 in the LCS if and only if, d ðqn ; Ls Þ ! qnþd , d ðqn ; Lp Þ ! qnþe , sider the factors correctness, sequence and response time. They 379
are of higher importance to logic than the other factors 380
323 where s \ p ¼ ;, d\e ¼ ; and d > 0, e > 0. Let us consider
because they have been the reason for the failure in the previ- 381
324 two parallel services possessing the states Qa and Qb respec-
ous similar incidents. Hence they are rated with highest impact 382
325 tively and the parallel execution is observed between them if
value 5. These impact values can also be given by the analyst 383
326 and only if, 9n dðqn ; Ls Þ ! qnþd \ d qn ; Lp !
according to his business and the prediction based on proba- 384
327 qnþe \ d qnþd ; Ls – > qnþe \ d qnþe ; Lp – > qnþd g where bility estimation. This is shown in step 9 of Fig. 2. 385
328 qn+d, qn+e qn+d 2 Qa and qn+e 2 Qb. qn+d should not reach For every impact value (1–5), only a particular percentage 386
329 any of the states in set Qb and qn+e should not reach any of the of allowance is allowed to act on the logic while making a 387
330 states in set Qa. For any service n iteration is said to be change. For example, if a decidability factor’s impact value 388
331 observed if and only if 9n d ðqn ; LÞ ¼> qn i.e. if any state qn is 5, then it is of higher impact i.e. the factor has been the 389
20 November 2015
as IL and impact value 1 is calculated together Impact_Very_- 410

Table 2 Impact values of the decidability factors.
Low abbreviated as IVL based on the formulation. The impact 411
Decidability factor Factor type Impact value from values determined are compared with the threshold values and 412
incidents if all the change, constraint and QoS factor values are within 413
Order of execution, IOOE Change 3 the limit of allowance indicated by the successfully satisfied 414
factor change, constraint and QoS factor patterns, then that particu- 415
Similarity measure, ISM Change 2 lar change request’s decidability changes to ‘1’. 416
factor Else, the decidability is set to ‘0’. The decidability algorithm 417
Business Policy Change 4 is given in Fig. 3. The algorithm adumbrates the decidability 418
Enforcement, IBPE factor
calibration by analyzing the impact analysis set I. Every decid- 419
Correctness, ICS Change 5
factor
ability factor value CFV is compared with the maximum suc- 420
Synchronization, ISY Constraint 4 cessful decidability factor values CFM and if found greater 421
factor that the maximum value, is checked for its corresponding 422
Branch, IBR Constraint 3 impact factor. The threshold values are computed with respect 423
Factor to the impact factor values. For instance, threshold high is 424
Parallel, IPL Constraint 2 computed if the impact factor belongs to impact high. Based 425
factor on this decidability is calibrated. 426
Sequence, ISQ Constraint 5 When the decidability is changed to ‘1’, the various cellular 427
factor automaton patterns are generated based on this decidability 428
Iteration, IIT Constraint 1
and manageability for rules, functions and parameters as 429
factor
Availability, IAV QoS factor 2
shown in step 11 of Fig. 2 and the approval is sent to the run- 430
Execution time, IET QoS factor 4 time manager. 431
Response time, IRT QoS factor 5

Cost, ICT QoS factor 3 3.1. Estimation of probability through incident matching 432
The various cellular automata patterns including the property, 433

change, request, QoS, rule, function, parameter, and incident 434
390 reason for the failure of a change in the maximum number of
patterns generated act as the knowledge base to aid in knowl- 435
391 previous incidents and hence only 40% of its impact is allowed
edge discovery and decision making. The probability of risk, 436
392 to act on the logic where as if a decidability factor’s impact
accuracy, automation, and incident matching is determined 437
393 value is 1, it is of lower impact hence 60% of its impact is
and from the extracted knowledge, the functional deviations, 438
394 allowed to act on the logic. This allowance of impact is termed
variations in QoS, violations in SLA and policy and failure 439
395 as the threshold in the impact analysis part. The threshold cal-
are discovered. Probability of accuracy indicates the extent 440
396 culation for each kind of impact value is listed in Table 3. This
to which the evaluations are accurate. This is estimated 441
397 is shown in step 10 of Fig. 2. Threshold estimation, Ti is based
based on the following formulation, PðaccuracyÞ ¼ 442
398 on the following: Ti ¼ Df ð1 þ IAÞ, where, Df = maximum Pn Pn Pn
399 value of the decidability factor from successful incidents, x¼1 ConEvalðxÞ þ x¼1 ChangeEvalðxÞ þ x¼1 QoS Eval 443
400 IA = percentage of impact allowed for that threshold type i. ðxÞ=ð3 nÞ. 444
401 As the impact values are rated based their effect on logics, where ConEval(x) is the result of evaluation of the con- 445
402 the impact factor determination is also classified according to straint factors for incident x by the cellular automata, 446
403 the rating. Hence the impact factor for the decidability factors ChangeEval(x) is the result of evaluation of the constraint fac- 447
404 having impact value 5 is calculated together Impact_Very_ tors for incident x by the cellular automata, QosEval(x) is the 448
405 High abbreviated as IVH, impact value 5 is calculated together result of evaluation of the constraint factors for incident x by 449
406 Impact_Very_High abbreviated as IVH, impact value 4 is cal- the cellular automata and n is the total number of incidents 450
407 culated together Impact _High abbreviated as IH, impact value recorded. Probability of risk indicates the amount of risk 451
408 3 is calculated together Impact_Medium abbreviated as IM, involved in the evaluations made. This is estimated based on 452
409 impact value 2 is calculated together Impact_Low abbreviated the following formulation, 453
( "P 454
X 5
i¼1 ConResi ðxÞ ConPri ðxÞ
n
PðriskÞ ¼ 1 P5
x¼1 i¼1 ConPri ðxÞ
P4
Table 3 Threshold values for impact values. ChangeResi ðxÞ ChangePri ðxÞ
þ i¼1 P4
Threshold name Impact Percentage of impact i¼1 ChangePri ðxÞ
value allowed (%) #,
X n
Threshold_Very_High 5 40 þ QoSEvalðxÞ ð3 nÞ
x¼1 456
– TVH
Threshold_High – TH 4 50
Threshold_Medium – 3 50
where ConResi(x) is the value of the constraint factor i in inci- 457
TM dent x in the cellular automata, ChangeResi(x) is the value of 458
Threshold_Low – TL 2 55 the change factor i in incident x in the cellular automata, 459

Threshold_Very_Low – 1 60 ConPri(x) is the priority for the constraint factor i of incident 460
TVL x, QosEval(x) is the result of evaluation of the constraint fac- 461
tors for incident x by the cellular automata, Change Pri(x) is 462
20 November 2015
Figure 3 Decidability calibration algorithm.
463 the priority for the constraint factor i of incident x, Probability Probability of incident matching is the estimation of the extent 473
464 of degree of automation is the estimation of the extent to to which incidents are matched. This is estimated based on the 474
465 which correct and smooth changes are made by automated following formulation, 475
Pn
x¼1 ½PropEvalðxÞ ConEvalðxÞ ChangeEvalðxÞ QoSEvalðxÞ
Pðdegree of automationÞ ¼
n
P 476
466 decisions made. ½ðPRC PRP Þ þ ðPSC PSP Þ
Pðincident matchingÞ ¼
467 where ConEval(x) is the result of evaluation of the constraint PRP þ PSP 478
468 factors for incident x by the cellular automata, ChangeEval(x)
469 is the result of evaluation of the constraint factors for incident
where PRC is the possible rule combinations for received rule, 479
470 x by the cellular automata, QosEval(x) is the result of evalua-
PSC is the possible service combinations for received composi- 480
471 tion of the constraint factors for incident x by the cellular
tion set, PRP existing rule combinations for received rule, PSP 481
472 automata and n is the total number of incidents recorded.
existing service combinations for received composition set. 482
20 November 2015
Figure 4 Algorithm for behavioral analysis.
483 This is followed by behavioral analysis using Probabilistic Cel- pared with the impact values configured for the cells, O1, O2, 519
484 lular Automata. O3, O4, at the respective positions. The transition rule applied 520
for the PCA is shown below: 521
485 3.2. Behavioral analysis using Probabilistic Cellular Automata 522
486 The Probabilistic Cellular Automata is a stochastic cellular Transition Rule for PCA:
487 automaton where two rules are followed synchronously for If (XN > O3 && XN < O1) then
488 the updation of the cells. One of the rules applied here is for Move cell position of XN to X2 (right)
489 the determination of the decidability of the automaton by End If
490 comparing every cell state with the threshold. The second rule If (XN > O4 && XN < O2) then
491 is for the determination of the cell positions dynamically based Move cell position of XN to X3 (bottom)
492 on the probabilistic measures. So, the states of the new entities End If
If (XN < O3) then
493 are chosen according to some probabilistic distribution and
Move cell position of XN to X4 (left)
494 behavioral analysis is thus performed and the predictions are End If
495 made. The algorithm for behavioral analysis is shown in If (XN > O2) then
496 Fig. 4. The algorithm adumbrates the dynamic change of cell Move cell position of XN to X1 (top)
497 positions based on the probability measures. Considering all End If
498 change requests, CR, the probability of risk is estimated from 525
499 the recorded incidents and is compared with the threshold 524
500 value which in turn is estimated through incident matching. The transition rule is elucidated in the following Fig. 5. On 526
501 If the probability value exceeds the threshold, then it indicates backtracking from cell X4, the corresponding FSM before 527
502 majority of the recorded incidents have failed to satisfy the and after change is extracted and this makes it possible to trace 528
503 decidability criterion. So, the impact ratings, priorities and back any point of change. 529
504 the cell positions of the factors have to be revised. Therefore, The change history stores the incident property, change fac- 530
505 the change requests with decidability failed, CD’ are extracted tor, constraint factor and QoS factor patterns and the audit 531
506 and the total sum of the difference between threshold and log keeps track of the change incidents, change outcome and 532
507 decidability factor value of all unsuccessful incidents is esti- the change log. From this rich knowledge base, the agent per- 533
508 mated for every decidability factor i. Then the decidability fac- forms decision making and risk analysis (Tjoa et al., 2011; 534
509 tor with greater deviation from the threshold is chosen and the Marschall et al., 2012). The patterns are detected and the Prob- 535
510 impact priority of the factor is increased. The decidability is abilistic Cellular Automata performs the behavioral analysis 536
511 verified and the cell positions are dynamically and syn- by estimating the probability of accuracy, risk, automation 537
512 chronously changed by the Probabilistic Cellular Automata and incident matching (Plebani and Pernici, 2009) by employ- 538
513 according to the transition rule as shown in step 12 of Fig. 2. ing the JESS rule engine for the purpose of decision making 539
and this ultimately leads to the goal and value prediction. 540
514 3.2.1. Transition rule
515 The transition rule followed by the PCA for changing the cell 4. Experimental results 541
516 positions dynamically is elucidated with the aid of Fig. 5. The

517 impact values at the top position, X1, at the right position X2, This section elucidates the change management scenario with 542
518 at the bottom position X3, and at the left position X4 are com- the help of an application from a business domain. Further, 543
20 November 2015
TOP
Change of cell positions based on
Probability of risk
X1
X4 >O1 X1<O1
X1<O3
X2>O2
X3>O2
X4>O3&& X4 <O1
RIGHT X4 X2 LEFT
X2<O3
X4>O4&& X4 <O2 X3<O3&& X3<O1

X1< O 2 && X1 >O4
X2<O2
X3<O3
X3
BOTTOM
S1
Before Change
R11
I/P: user id
I/P: user id
O/P:o1,o2,o3
O/P:o1,o2,o3,o4,o5,o6
O/P:o1,o2,o3
I/P: user id
R12 R21 R31

I/P:i1,i2,i3,i4,i5,i6 O/P: status O/P: status
S1 S2 S3
S1
After Change
R11
I/P: user id
I/P: user id
O/P:o1,o2,o3
O/P:o1,o2,o3,o4,o5,o6
O/P:o1,o2,o3
I/P: user id
R12 R21 R41

I/P:i1,i2,i3,i4,i5,i6 O/P: status O/P: status
S1 S2 S4
Figure 5 Rule followed for cell positioning and behavioral analysis.
544 the results of change impact analysis and incident matching change impact analysis framework extracts the corresponding 559
545 approach over the LCS have been depicted. source code based on the domain and context analysis as 560
shown in Fig. 6. After this, dependency analysis is done and 561
546 4.1. Change management scenario the business logic set is formed. The corresponding composed 562
partial schema before change is then built and validated with 563
547 In this section, an application from the travel domain is illus- the complete composed schema. 564
548 trated for elucidating the change management and change The schema under consideration is termed as a partial 565
549 impact analysis process. Consider a travel agency LCS that schema because it reflects only the part of the business logic 566
550 offers many types of functionalities airline, hotel and cab ser- under change while the complete composed schema reflects 567
551 vice. The composed service offers submission of the customer the entire business process. After successful schema validation, 568
552 details – get_cust_details, booking the air ticket – book_air- the Finite State Machine before change is constructed as 569
553 ticket, booking the hotel – book_hotel and booking the cab shown in Fig. 7. The property pre evaluation is performed over 570
554 – book_cab executing in sequence. this FSM and the results are shown in Fig. 8 calibrating the 571
555 Now, consider a change request which demands the manageability. 572
556 removal of ‘book_airticket’. This removal has to be done very On successful manageability calibration, the business ana- 573
557 cautiously since the execution of the other rules and services lyst is allowed to make a change. There are various criteria 574
558 must not be affected. On receiving a change request, the considered before the deletion process. A rule in the FSM can- 575
20 November 2015
Source Code Extracon
Travel-LCS
AirLine_Service Hotel_Service Cab_Service
@WebService(serviceName = @WebService(serviceName = “Taj @WebService(serviceName =

"Singapore_Airlines") Hotel") "Cab_Service")
ResultSet rs1=stmt.executeQuery code1 code1 String sql=("insert into code2 code2 String sql=("insert into
("Select * from airline where hotel(username,contact,address,roo taxi(username,contact,address,area,
username='"+uname+"'"); mtype,no_of_mem)" + " values" + type) values("+cust.get(0).toString()
while(rs1.next()) { "("+cust.get(0).toString() +","+cust.get(1).toString()+","+cust.g
cust.add(rs1.getString(2)); +","+cust.get(1).toString()+","+cust.ge et(2).toString()+","+area+","+type+")"
cust.add(rs1.getString(3)); t(2).toString()+“”); );
cust.add(rs1.getString(4)); }
Schema Generaon
<LCS Schema [Airline][Hotel][Cab]>

<cservice>
- <service name="Singapore_Airlines">
- <rule id="r11" name="getcust_details">
- <input>
<parameter name="uname" ptype="String" />
</input>
</rule>
- <rule id="r12" name="book_airticket">
<call rule="getcust_details" service="Singapore_Airlines" />
- <input>
<parameter name="contact" ptype="String" />
<parameter name="address" ptype="String" />
<parameter name="source" ptype="String" />
<parameter name="dest" ptype="String" />
<parameter name="toc" ptype="String" />
</input>
- <output>
<rtype>String</rtype>
</output>
</rule>
Figure 6 Source code extraction, schema generation and business logic set extraction for travel LCS.
576 not be deleted if it is being called by any other rule or if it is the and the probabilistic measures are estimated for behavioral 598
577 only call to any other rule or if there is no link between the pre- analysis. PCA then performs the dynamic updation of the cell 599
578 vious and the next rule or service other than the rule to be states followed by the decidability calibration. The change is 600
579 deleted. Only then, the analyst is allowed to make the initial then committed on successful decidability calibration and 601
580 move toward deletion. This aids in reducing the risk involved rolled back if the calibration fails. The impact analysis done 602
581 in making the change in addition to the change evaluation pro- based on the estimated decidability factor values is shown in 603
582 cess. ‘book_airticket’ is neither called by any other rule nor is Fig. 11. Probability of risk is estimated as P (risk) = 0.233. 604
583 the only call to any other rule. Probability of accuracy is estimated as P (accuracy) = 605
584 It is not the only link between the previous and the next rule 0.7667, Probability of degree of automation as 606
585 either. So, the analyst is allowed to make the change at the P (aut) = 0.723 and probability of incident matching as 607
586 schema level. The constructed FSM after the incorporation P (inc) = 0.89. Fig. 12 shows the results of impact analysis 608
587 of the deletion of ‘book_airticket’ is shown in Fig. 9. Fig. 10 for 50 change requests to travel LCS for different impact fac- 609
588 depicts the change process and the schema and FSM before tors. The blue curve indicates the decidability factor value of 610
589 and after change. the particular factor belonging to the corresponding impact 611
590 Before the change is committed, the change factor evalua- factor category and the red curve indicates the corresponding 612
591 tion, constraint factor evaluation and QoS factor evaluation threshold estimations. 613
592 for the travel agency application are performed and the corre- Fig. 13 shows the change in cell position based on the prob- 614
593 sponding decidability factor values are estimated. This is ability function from previous neighboring states at different 615
594 followed by change impact analysis where the previously time stamps and for different change requests. Figs. 14–17 616
595 recorded similar change incidents are extracted and the impact show the improvement observed in the response time, execu- 617
596 value estimations are done as shown in Fig. 11. The threshold tion time, cost and availability after change is made with 618
597 estimations and the impact factor estimations are also done change impact analysis. The blue curve indicates the estima- 619
20 November 2015
Figure 7 Constructed FSM before change for travel LCS.
Property Evaluaon
Figure 8 Property evaluation results for travel LCS.
20 November 2015
Figure 9 Constructed FSM after change for travel LCS.
Change Evaluaon
Before Change Aer Change
<LCS Schema [Airline][Hotel][Cab]> <LCS Schema [Airline][Hotel][Cab]>

<cservice> <cservice>
- <service name="Singapore_Airlines"> - <service name="Singapore_Airlines">
- <rule id="r11" name="getcust_details"> - <rule id="r11" name="getcust_details">
- <input> - <input>
<parameter name="uname" ptype="String" /> <parameter name="uname" ptype="String" />
</input> </input>
</rule> </rule>
- <rule id="r12" name="book_airticket"> </service>
<call rule="getcust_details" service="Singapore_Airlines" />
- <input>
<parameter name="contact" ptype="String" />
<parameter name="address" ptype="String" />
<parameter name="source" ptype="String" />
<parameter name="dest" ptype="String" />
<parameter name="toc" ptype="String" />
/i
Figure 10 Change evaluation for travel LCS.
20 November 2015
Figure 11 Impact analysis results for 50 change requests and impact succeeded change requests.
620 tion before change and red curve indicates the estimation after made i.e. how accurate is the behavior of the framework. This 648
621 change. is estimated from the number of manageability and decidabil- 649
ity automatons which have succeeded in incorporating the 650
change. 651
622 4.2. Goal formulation
5. Related work 652
623 The various goals of the framework have been identified and
624 depicted in Table 4. The proposed framework aims at provid-
625 ing efficient change management with reduced risk, high accu- In this section the various works pertinent to the proposed 653
626 racy, high degree of automation and incident matching. These framework and approaches are discussed. Change manage- 654
627 goals of the framework are evaluated based on the metrics ment is a structured approach for making sure that the changes 655
628 shown in the table. For the estimation of risk, the number of are thoroughly and smoothly made and that the lasting bene- 656
629 automatons in which either manageability i.e. property evalu- fits of the change are achieved. 657
630 ation or decidability i.e. change evaluation comprising of

631 change factor, constraint factor and QoS factor evaluation 5.1. Related change management work 658
632 has failed is to be extracted from the entire set of incidents.

633 This indicates the extent to which runtime support and deci- Liu et al. (2011) have presented a framework where managing 659
634 sion making assured by the framework have failed to meet changes in LCSs has been modeled as a dual service query 660
635 the business needs. optimization process (Liu et al., 2011). The optimization has 661
636 The degree of automation is the extent to which the deci- been performed by considering only the non functional factors 662
637 sions are automatically made by the framework. This is indi- like reputation and the QoS factors. However, we have consid- 663
638 cated by the number automatons which have proceeded to ered the factors for enabling the analysts to manage the func- 664
639 and succeeded in the decidability calibration after manageabil- tional and non functional requirements of the consumers. 665
640 ity calibration from the entire set of incidents i.e. the extent to Akram et al. (2010) have proposed an automatic change 666
641 which the decisions have been automatically made. Incident management framework that is based on the petri net models 667
642 matching indicates the possible number of combinations of which is used to manage triggering changes and reactive 668
643 rule and service patterns out of the existing number of patterns changes (Akram et al., 2010). Also the functional and non- 669
644 from the entire set of incidents i.e. the depth and level of inci- functional properties of the composed services have been iden- 670
645 dent matching performed by the framework. Accuracy is the tified. However, we have used FSM for the modeling of the 671
646 measure which indicates the extent to which the decisions changes as they facilitate incident matching with low time 672
647 made by the framework have been correctly and cautiously and space complexity. We also have identified the QoS 673
20 November 2015
Figure 12 Impact analysis results (a) for Impact_VeryHigh, (b) for Impact_High, (c) for Impact_Medium and (d) for Impact_Low.
674 parameters into account as it has to satisfy the service provi- 2013). But there is no methodology adopted for performing 693
675 ders and analysts since as business policies change continu- emergency changes and change impact analysis was not done. 694
676 ously, there is a need for service-oriented systems to be In our work, by implementing Finite State Machine, a stan- 695
677 submissive and adaptive. Dijkman et al. (2011) have proposed dard methodology is adopted for the evaluation of manage- 696
678 the metrics for the evaluation of the node matching, structural ability and decidability factors. Craiu and Lee (2006) have 697
679 and behavioral similarities of business (Rajeswar et al., 2014). developed likelihood-based methods for estimating rules of 698
680 The evaluation of similarity among business process models is cellular automata aimed at the regeneration of observed regu- 699
681 only dealt whereas its applicability in web service change man- lar patterns (Craiu and Lee, 2006). But the idea of pattern gen- 700
682 agement is not considered. Tjoa et al. (2011) have proffered a eration using likelihood inference has not been focused in the 701
683 formal approach which enables risk-aware business process view of change management. In our work, we have imple- 702
684 modeling and simulation (Koriem et al., 2003). But the estima- mented Probabilistic Cellular Automata (PCA) for the infer- 703
685 tion of risk is based on the detection of intrusion and the esti- ence from historical patterns of change incidents. The 704
686 mation of policy violations alone. In our work, we have concept of Probabilistic Arithmetic Automata (PAA) 705
687 estimated the metric risk from the evaluation of change, con- described as markov chains over a larger state space and their 706
688 straint and QoS factor evaluation. benefits in their utility as a modeling technique has been con- 707
templated by Marschall et al. (2012) and Dijkman et al. (2011). 708
The claim of PAA as a modeling technique and its justification 709
689 5.2. Methodology adopted in the related work
has paved the way for the idea of implementing the Probabilis- 710
tic Cellular Automata (PCA) in our work for the decision 711
690 Liu et al. (2013) have proposed a change management frame- making and prediction from the historical change incidents. 712
691 work where managing changes in LCS which deals with both Maamar et al. (2008) have suggested a holistic approach for 713
692 functional and non functional change requests (Liu et al., managing software change impact (Yu et al., 2008). But, this 714
20 November 2015
Figure 13 Cell position change based on probability function and behavioral analysis using PCA.
Figure 14 Change in the estimation of response time for 100 Figure 16 Change in the estimation of cost for 100 change
change requests of travel LCS. requests of travel LCS.
Figure 15 Change in the estimation of execution time for 100 Figure 17 Change in the estimation of availability for 100
change requests of travel LCS. change requests of travel LCS.
20 November 2015
QoS parameters of various services, but this work does not 732
Table 4 Evaluation metrics for CIA framework.
consider the various trade-offs that will be faced by QoS based 733
Goals Formulation composition during runtime. These QoS factors have not been 734
Risk P hm0 hd0 ii focused in the change management view of web services. Thus 735
RiskðRIÞ ¼ m [ d 100
in our work, QoS factors are considered during the change 736
m0 – Number of automata in which
management phase, as they help in verifying the impact caused 737
manageability failed
m – Total number of automata generated in
due to the change over the quality of the business process. The 738
manageability pattern cookie (Venkatesan et al., 2013) based accounting model pro- 739
d0 – Number of automata in which decidability posed identifies duplicate service requests which are used by 740
failed most of the attackers to affect the availability of the servers. 741
d – Total number of automata generated in This model works by recording each request in the client side 742
decidability pattern

and a hash value of the corresponding request is also stored 743
Degree of P ½1mm0 \½1dd0 in the server. Thus any client attempting to send a duplicate 744
Degree of Automation ¼ CRn 100
automation request or modify the content of the cookie can be easily iden- 745
m0 – Number of automaton in which tified, thereby protecting the server from DOS attacks. 746
manageability failed
m – Total number of automatons generated in
manageability pattern 5.4. Related work based on petri nets 747
d0 – Number of automaton in which decidability
failed Koriem et al. (2003) has proposed a system called SRN Equiv- 748
d – Total number of automaton generated in alent Model (EM) for evaluating the performance of POS 749
decidability pattern
based system. This model is a combination of petri net theory 750
CRn – Number of change requests
P and control theory and has two main nets. The main net con- 751
Incident Patterns ¼ ½ðPRP CPRC þ PSP CPRC Þ
matching PRC – Possible rule combinations for received sists of the known parts and the other net (second) consists of 752
rule parts that feed the known parts. Considering petri nets for 753
PFC – Possible service combinations for received change management has certain difficulties as their complex 754
composition structures brings difficulties in the analyzing them. Moreover, 755
PRP – Existing rule combinations for received representation of petri nets is task based which is not suitable 756
rule for analysts. Therefore, in our work FSM representation of the 757
PFP – Existing service combinations for received services in terms of rules, policy etc. make it easier for the ana- 758
composition
Accuracy P h 0
h 0
ii lyst to achieve successful changes. FSM also achieves the 759
Accuracy ¼ 1 mm \ 1 dd 100 decidability of the problems that were not possible when petri 760
m0 – Number of automaton in which nets were used. The verification and validation of the changes 761
manageability failed will also become a complex task as the number of services 762
m – Total number of automatons generated in increases. 763
manageability pattern
d0 – Number of automaton in which decidability
failed 6. Discussion 764
d – Total number of automaton generated in

decidability pattern In this section, we present the contributions of our research 765
and its limitations. 766
6.1. Research contributions 767

715 work illustrates how context and policy are woven into web
716 services composition scenarios only. In our work, we have This research makes the following contributions by presenting 768
717 incorporated the estimation of Business Policy Enforcement a formal approach for change impact analysis of LCS using 769
718 (change factor) for hunting for any policy violations among Probabilistic Cellular Automata. Specifically this research: 770
719 services and accessibility (property) for checking if the services
720 in composition are accessible. All these issues are the hurdles in Presents a formal approach for analyzing the LCS for 771
721 the current scenario which hinder the change management pro- effective runtime management by property evaluation and 772
722 cess in long term composed services in enabling the analysts to to evaluate the changes over the LCS by change 773
723 make the changes in a timely and cost effective manner and evaluation using Finite State Machine as the 774
724 with reduced risk. methodology. 775
Presents a novel framework with standard procedures and 776
725 5.3. Related works in web services methodologies for managing the changes in a timely and 777
cost effective manner finding the inconsistencies in the 778
726 The analysis done (Rajeswar et al., 2014) brings out the impor- changes made. 779
727 tance of considering QoS for service composition. This Presents and performs change impact analysis and behavior 780
728 research work has analyzed a number of QoS requirements, analysis over the knowledge extracted from the patterns 781
729 various QoS models and also the difficulties that are faced using Probabilistic Cellular Automata (PCA) with efficient 782
730 for QoS based service composition. The research work has incident matching enabling to trace the exact reason for 783
731 provided a utility function that can be used to evaluate the failure in case of change being discarded. 784
20 November 2015
785
786 In the existing works, petri nets are used as mathematical sent the safe and unsafe states of the Service Oriented Enter- 845
787 models which help in depicting the flow of a particular process prises (Akram et al., 2010). When a new change request 846
788 (Tan et al., 2009). A petri net structure has 4 components – arises for a new LCS set which has never been encountered 847
789 state, transition, token and edge. Petri nets are complex to before, the change impact analysis and incident matching are 848
790 be analyzed and moreover the number of states increases to not done. Though the property, change, constraint and QoS 849
791 a very large extent when the number of services in composition factor evaluation are evaluated, the inference cannot be made 850
792 is large. Petri nets are suitable for concurrent and complex pro- based on the probabilistic measures from the previous inci- 851
793 cesses like business process alignment, workflow management dents. However the results of the new change are recorded as 852
794 and the focus is more on the top down changes alone (Tan a new incident and this aids in impact analysis for future sim- 853
795 et al., 2009) whereas Finite State Machines are suitable for ilar change requests. 854
796 sequential process with every state followed by a transition Change management provides large space for intense 855
797 to another state or end state which is the focus of our work research. Future enhancements of this work would be to imple- 856
798 (Chenthati et al., 2010). The petri nets are represented as tasks ment tuning machines in order to support analysis of concur- 857
799 [Event, Action. Condition] and are suitable for event based rent processes extending the framework to support change 858
800 change verification which is suitable only for the IT developers optimization and to support automated identification of 859
801 and not for the analysts, whereas FSM which is represented as change decay over time. 860
802 rule, function, and policy assures a representation supporting

803 the analysts which is much needed in the change scenario. 7. Conclusion 861
804 Decidability of the problems is not possible to be judged using
805 petri nets whereas it is possible using FSM with standard algo- Based on the behavioral analysis, the degree of automation, 862
806 rithmic conversion and automated decision making. The accuracy, incident matching and risk are to be predicted based 863
807 default language handling capability of the FSM notifies every on incident matching. Hence the proposed framework is 864
808 new member added to the language, so the analyst can be noti- expected to assess, change and manage the business logic of 865
809 fied when any new rule, function or even a parameter is added the web services in long term composition with efficient inci- 866
810 to the logic which is not possible in petri nets. FSM is the best dent matching, high degree of automation and authorization, 867
811 choice in case of incident mapping because of its backtracking high degree of knowledge transfer, manageability, decidability, 868
812 capability with cellular automata compared to petri nets. Rule high success and incident rates at low risk and cost without 869
813 level changes in the reflected directly in the source code by the compromising the quality of the business process. The frame- 870
814 FSM. work facilitates in choosing the ‘best’ services from multiple 871
815 With respect to the behavioral analysis, petri nets can be service providers who compete to offer the similar services with 872
816 used to model the behavior of concurrent process whereas a different user-centric quality and paves way for the con- 873
817 FSM can be used to model the I/O behavior which is much sumers or end users to be highly benefited from the open com- 874
818 needed in our case. The primary goal of the framework is to petition among businesses. This aids in the enhancement of the 875
819 respond to the changes correctly and cautiously at low cost change management process in long term composed services so 876
820 and time. Since petri nets are complex structures they demand that the changes can be made by the analysts in a timely and 877
821 large memory conception and high processing time leading to cost effective manner and with reduced risk. 878
822 high time and space complexity. Petri nets are comparatively
823 slow and inefficient in the retrieval of services and inflexible 8. Uncited references 879
824 in searching and tracking of them. There are high changes
825 for a petri net based system to reach an undesirable state
826 due to the loss of its token. Hence, petri nets are suitable only Graves et al. (2000), Malak et al. (2009), Wu et al. (2009). 880
827 for modeling concurrent processing associated with determin-

828 istic task, event and condition and not for modeling non deter- References 881
829 ministic sequential processing associated with language theory

830 like FSM. Akram, Salman, Bouguettaya, Athman, Liu, Xumin, Haller, Armin, 882
Rosenberg, Florian, Xiaobing, Wu, 2010. Change management 883
framework for service oriented enterprises. Int. J. Next-Generation 884
831 6.2. Limitations
Computing (IJNGC) 1 (1), 1–077. 885
Apostolou, Dimitris, Mentzas, Gregoris, Stojanovic, Ljiljana, Thoens- 886
832 Generalization of the results from this work must be done with sen, Barbara, Lobo, Tomás Pariente, 2010. A collaborative 887
833 caution in light of its limitations. Though petri nets have a lot decision framework for managing changes in e-Government 888
834 if issues with respect to our focus as mentioned, they also have services. Elsevier J. Gov. Inf. Q. 28 (1), 101–116. 889
835 a few advantages compared to FSM. The representation as Chenthati, Deepak, Vaddi, Supriya, Mohanty, Hrushikesha, Damo- 890
836 tasks enables high expressiveness in petri nets while the expres- daram, Avula, 2010. Verification of web services modeled as finite 891
state machines. In: Proceedings of IEEE International Conference 892
837 siveness is low in the case of FSM. Petri nets are much suited
on Mathematical/Analytical Modeling and Computer Simulation 893
838 for concurrent processes and in the case of the change scenario (AMS), pp. 526–531. 894
839 demanding the analysis of concurrent processes, petri nets are Chua, David K.H., Aslam Hossain, Md., 2012. Predicting change 895
840 more advantageous (Akram et al., 2010; Tan et al., 2009). propagation and impact on design schedule due to external 896
841 Though this problem can be solved when FSM are used by changes. IEEE Trans. Eng. Manage. 59 (3), 483–493. 897
842 the use of Tuning Machines, petri nets are readily suitable Craiu, Radu V., Lee, Thomas C.M., 2006. Pattern generation using 898
843 for concurrent deterministic processing while FSM are not likelihood inference for cellular automata. IEEE Trans. Image 899
844 suitable for concurrent processing. Petri nets can easily repre- Process. 15 (7), 1718–1727. 900
20 November 2015
901 Dijkman, Remco, Dumas, Marlon, van Dongen, Boudewijn, Kaarik, ysis on various web service composition approaches based on QoS 937
902 Reina, Mendling, Jan, 2011. Similarity of business process models: factors. J. King Saud Univ. Comput. Inf. Sci. 26 (1), 143–152. 938
903 metrics and evaluation. Elsevier J. Inf. Syst. 36 (2), 498–516. Rovegard, Per, Angelis, Lefteris, Wohlin, Claes, 2008. An empirical 939
904 Graves, Todd L., Karr, Alan F., Marron, J.S., Siy, Harvey, 2000. study on views of importance change impact analysis issues. IEEE 940
905 Predicting fault incidence using software change history. IEEE Trans. Software Eng. 34 (4), 516–530. 941
906 Trans. Software Eng. 26 (7), 653–661. Setzer, Thomas, Bhattacharya, Kamal, Ludwig, Heiko, 2010. Change 942
907 Koriem, Samir M., Dabbous, T.E., El-Kilani, W.S., 2003. A new petri scheduling based on business impact analysis of change-related 943
908 net modeling technique for the performance analysis of discrete risk. IEEE Trans. Network Serv. Manage. 7 (1), 58–71. 944
909 event dynamic systems. J. King Saud Univ. Comput. Inf. Sci. 15, Tan, Wei, Fan, Yushun, Zhou, MengChu, 2009. A petri net-based 945
910 95–128. method for compatibility analysis and composition of web services 946
911 Liu, Xumin, Bouguettaya, Athman, Qi, Yu, Malik, Zaki, 2011. in business process execution language. IEEE Trans. Autom. Sci. 947
912 Efficient change management in long-term composed services. Eng. 6 (1), 94–106. 948
913 Springer J. Serv. Oriented Comput. Appl. 5 (2), 87–103. Tjoa, Simon, Jakoubi, Stefan, Goluch, Gernot, Kitzler, Gerhard, 949
914 Liu, Xumin, Akram, Salman, Bouguettaya, Athman, 2011. Change Goluch, Sigrun, Quirchmayr, Gerald, 2011. A formal approach 950
915 Management for Semantic Web Services, first ed. Springer Science enabling risk-aware business process modeling and simulation. 951
916 and Business Media, ISBN 978-1-4419-9328-1. IEEE Trans. Serv. Comput. 4 (2), 153–166. 952
917 Liu, Xumin, Bouguettaya, Athman, Wu, Jemma, Zhou, Li, 2013. Ev- Venkatesan, S., Saleem Basha, M.S., Chellappan, C., Vaish, Anurika, 953
918 LCS: a system for the evolution of long-term composed services. Dhavachelvan, P., 2013. Analysis of accounting models for the 954
919 IEEE Trans. Serv. Comput. 6 (1), 102–115. detection of duplicate requests in web services. J. King Saud Univ. 955
920 Maamar, Zakaria, Benslimane, Djamal, Mostéfaoui, Ghita Kouadri, Comput. Inf. Sci. 25 (1), 7–24. 956
921 Subramanian, Sattanathan, Mahmoud, Qusay H., 2008. Toward Wang, Qianxiang, Shao, Jin, Deng, Fang, Liu, Yonggang, Li, Min, 957
922 behavioral web services using policies. IEEE Trans. Syst. Man Han, Jun, Mei, Hong, 2009. An online monitoring approach for 958
923 Cybern. 38 (6), 1312–1324. web service requirements. IEEE Trans. Serv. Comput. 2 (4), 338– 959
924 Malak, J.S., Mohsenzadeh, M., Seyyedi, M.A., 2009. Multi agent 351. 960
925 based web service QoS management architecture. In: Proceedings Wu, Yu, Wu, Jie, Jun Wang, Jia, Zhang, Shiyong, 2009. A multi-agent 961
926 of the IEEE Conference on Computational Intelligence and assistant dynamic resource management model of composite 962
927 Software Engineering, pp. 280–286. service. In: Proceedings of the IEEE Conference on Computational 963
928 Marschall, Tobias, Herms, Inke, Kaltenbach, Hans-Michael, Rah- Intelligence and Software Engineering, pp. 01–04. 964
929 mann, Sven, 2012. Probabilistic arithmetic automata and their Xiong, Peng Cheng, Fan, Yu Shun, Zhou, Meng Chu, 2009. Web 965
930 applications. IEEE/ACM Trans. Comput. Biol. Bioinf. 9 (6), 1737– service configuration under multiple quality-of-service attributes. 966
931 1750. IEEE Trans. Autom. Sci. Eng. 6 (2), 311–321. 967
932 Plebani, Pierluigi, Pernici, Barbara, 2009. URBE: web service retrieval Yu, Qi, Liu, Xumin, Bouguettaya, Athman, 2008. Deploying and 968
933 based on similarity evaluation. IEEE Trans. Knowl. Data Eng. 21 managing web services: issues, solutions and directions. Springer 969
934 (11), 1629–1642. VLDB J. 17 (3), 537–572. 970
935 Rajeswar, M., Sambasivam, G., Balaji, N., Saleem Basha, M.S., 971
936 Vengattaraman, T., Dhavachelvan, P., 2014. Appraisal and anal-

A Formal Approach For Change Impact Analysis of Long Term Composed Services Using Probabilistic Cellular Automata

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

A Formal Approach For Change Impact Analysis of Long Term Composed Services Using Probabilistic Cellular Automata

Uploaded by

Copyright:

Available Formats

JKSUCI 206 No.

King Saud University

3 A formal approach for change impact analysis

7 Pondicherry Engineering College, Puducherry, India

8 Received 11 January 2014; revised 26 June 2014; accepted 23 October 2014

21 1. Introduction the partnership among the services in composition and the 28

Figure 1 Change impact analysis framework.

Property Before Aer FSM Construcon (before Business Logic Schema

Threshold name Impact Impact Threshold Decidabil- Name Type Impact

Figure 2 Workflow diagram of change impact analysis framework.

degree of made can be committed or not through the estimation of 364

as IL and impact value 1 is calculated together Impact_Very_- 410

Execution time, IET QoS factor 4 time manager. 431

Response time, IRT QoS factor 5

The various cellular automata patterns including the property, 433

TM dent x in the cellular automata, ChangeResi(x) is the value of 458

Threshold_Low – TL 2 55 the change factor i in incident x in the cellular automata, 459

Figure 3 Decidability calibration algorithm.

Figure 4 Algorithm for behavioral analysis.

516 positions dynamically is elucidated with the aid of Fig. 5. The

X4>O4&& X4 <O2 X3<O3&& X3<O1

R12 R21 R31

R12 R21 R41

Figure 5 Rule followed for cell positioning and behavioral analysis.

Source Code Extracon

AirLine_Service Hotel_Service Cab_Service

@WebService(serviceName = @WebService(serviceName = “Taj @WebService(serviceName =

<LCS Schema [Airline][Hotel][Cab]>

Figure 7 Constructed FSM before change for travel LCS.

Figure 8 Property evaluation results for travel LCS.

Figure 9 Constructed FSM after change for travel LCS.

<LCS Schema [Airline][Hotel][Cab]> <LCS Schema [Airline][Hotel][Cab]>

Figure 10 Change evaluation for travel LCS.

630 ation or decidability i.e. change evaluation comprising of

632 has failed is to be extracted from the entire set of incidents.

d – Total number of automaton generated in

6.1. Research contributions 767

802 rule, function, and policy assures a representation supporting

827 for modeling concurrent processing associated with determin-

829 ministic sequential processing associated with language theory

You might also like