Professional Documents
Culture Documents
19
maker into numerical values according to Saaty’s 1-9 Negative flows: expresses the weakness of ai
scale of pairwise comparisons shown in Table 2.
3- Step 3: Determines the relative importance of
Φ − (ai ) = ¦ P(ak ,ai )
a k ≠ ai
(5)
elements by calculating the corresponding maximum • Step 4: PROMETHEE II is used to rank alternatives
vectors values of the comparisons matrix. according to the score order of the net flows (ai)
4- Step 4: Checks the coherence of the subjectivity defined as follows:
values.
Φ (ai ) = Φ + (ai ) − Φ − (ai ) (6)
AHP method makes use of two measures: the Consistency
Index and the Coherence Ratio which are presented below.
3. Related work
Consistency Index (CI): it is obtained as follows:
λmax − n It is well-known that the Skyline operator is very useful for
CI = (1)
n −1 several applications that require a multi-criteria decision
Where max represents the maximum value corresponding making. It was introduced at the early 2000's in the field of
to the comparison matrix by pairs, n represents the number of databases by [12]. Many algorithms for computing the skyline
elements compared. have been proposed in the literature. BNL (Block Nested
Loops) algorithm makes a full scan of the dataset in several
Coherence Ratio (CR): it is computed as follows: iterations for computing the Skyline; for each iteration, it
CI keeps in memory the list of skyline candidates. D&C (Devide
CR = 100* (2) and Conquer) [12] method converses the set of points in a
ACI
group of partitions and calculates a result Skyline for every
Where ACI represents the average coherence indication, see partition, by using any Skyline algorithm; finally, it merges
[10] for more details. the results of partitions to find the final result [12]. [13]
A value of CR less than 10% is generally acceptable; proposed both Index and Bitmap-based algorithms which
otherwise, judgments must be reviewed and improved. allow returning Skyline result progressively. [14] introduced
algorithms NN (Nearest Neighbor) and BBS (Branch and
2.3. PROMETHEE Method Bound Skyline) based on R-trees indexing structure.
Recently, the problem of web services selection has
PROMETHEE is a MCDM method which allows making a received a considerable intention, several works were
partial or complete ranking of alternatives according to their proposed, adapting a variety of Skyline algorithms for service
pertinence, including the preferences of the user (weights, selection. [15] addresses several problems, notably, (i) taking
preference and indifference thresholds). into consideration users preferences in web services selection
PROMETHEE Method was proposed by [11], it is a part of and composition, optimization of service selection for several
outranking methods where alternatives are pair-wise users with conflicting preferences, (ii) service selection with a
compared. With PROMETHEE the notion of criterion is good compromise between attributes of QoS and (iii) dealing
extended including a function which expresses the preferences with the uncertainty aspect of QoS data during the process of
of the user for one alternative in opposition to another. Skyline services selection. [6] adapts existing Skyline
The steps of PROMETHEE method are: algorithms to calculate the best SEP (Service Execution Plan).
• Step 1: fixes for every criterion, one of six basic The result Skyline SEP, represents the set of SEPs that are not
types offered in PROMETHEE [11] (usual criterion, dominated by any other SEPs. [16] introduces a new concept
U-shape criterion, V-shape criterion, level criterion, called degree of dominance, which can capture the scale of
V-shape with indifference criterion and Gaussian dominance of Skyline services and order Skyline results, the
criterion). authors explain how the Skyline can help to determine the
• Step 2: For each couple of alternatives (ai, ak); the optimum allocation of weights to the different metrics of QoS,
preference degree (Outranking degree) is calculated relying on the dominance Skyline relation, the number of
in the following way: candidate services can be reduced and selection process can
¦
n then be simplified.
P(ai ,ak ) = π j .Fj (ai ,ak ) (3)
j=1 The approach to use MCDM methods to resolve some
Where j represents the weight of the jth criterion. problems is not new. In the context of web service selection
many methods are used such as AHP [7], PROMETHEE [17],
• Step 3: Calculates the negative and positive flows for
ELECTRE TRI [18] and MAUT [19]. However, we did not
each alternative ai:
find a lot of research works which combine two or several
Positive flows: expresses the force of alternative ai
MCDM methods to rank web services. In the work proposed
¦
+
Φ (ai ) = P(ai ,ak ) (4) by [8], the authors combine both methods ANP (Analytical
a k ≠ai Network Process) and PROMETHEE for ranking services,
while [20] and [21] combines AHP and PROMETHEE.
20
QoS Data history invocations
Decomposes the problem into a hierarchy
21
Algorithm 1. Ranking Skyline Services. meteorology like temperature, speed of wind, visibility,
meteorological conditions, etc.
The set of QoS metrics are: response time, throughput,
reliability, best practices, material and class.
1: Input: Skyline services (incomparable services) Gj (Si)
2: Input: Preference thresholds Pref [Qosi]
3: Input: Indifference thresholds Indef [Qosi]. Table 3. Examples of data of QoS [22].
4: Input: Preference Way: Sense [Qosi].
(%)
5: Input: Weight of criteria of Qos Weight [Qosi], result of AHP method.
Name
Time (Ms)
Response
(hits / dryness)
Throughput
(%)
Reliability
Material (%)
Class
Is Skyline
Best
6: output Skyline services ranked
7: //Computing the preference intensity
Practices
8: For I = 1 to n do
9 For k = 1 to n do
10 For j = 1 to m do
11 Diffj [Sk, Sj] = Gj (Si) - Gj [Sk];
12 If Diffj [Si, Sk] <0 and Sense [Qosi] = min then S1 FastWeather 125.44 13.5 86.4 80 91 4 Y
13 If Indef [Qosi] < ABS (Diff [Qosi]) <= Pref [Qosi] then
14 Fj (Si, Sk) = (ABS (Diffj [Sk, Sj]- Indef [Qosi]) / (Pref [Qosi] - S2 dotsfastweather 129.67 13.2 84.1 80 90 4 N
15 Indef [Qosi]);
16 else S3 WeatherForecast 261 1.8 58.1 80 94 3 Y
17 If ABS (Diffj [Sk, Sj]))> Pref [Qosi] then
18 Fj (Si, Sk) = 1 S4 WeatherFetcher 160 2.2 73.3 84 32 3 Y
19 else S5 WeatherService 190.5 12.4 54 80 8 2 N
20 Fj (Si, Sk) = 0
21 else S6 GlobalWeather 1463.5 2.4 53.5 84 42 2 Y
22 If Diffj [Sk, Sj]>0 and Sense [Qosi] = max then
23 If Indef [Qosi] <= Diffj [Sk, Sj] <= Pref [Qosi] then S7 ndfdXML 409.33 1.8 41.4 72 96 1 Y
24 Fj (Si, Sk) = (Diffj [Sk, Sj]- Indef [Qosi]) / (Pref [Qosi] -
25 Indef [Qosi]);
26 else 6.1. Skyline services computation
27 If Indef [Qosi]> Diffj [Sk, Sj] then
28 Fj (Si, Sk) = 0;
29 else In this demonstrative example, the two services S2 and S5
30 Fj (Si, Sk) = 1; are not in the Skyline result because S2 is dominated by S1 and
31 end for S5 is dominated by S1 too. The five services (S1, S3, S4, S6 and
32 end for
33 end for
S7) are not dominated by any other service, and they are
34 //Computing the preference degree : incomparable between them. The column “Is Skyline” of Table
35 For i = 1 to n do 3 indicates if a Web service is (or not) a skyline service.
36 For k = 1 to n do
37 nap = 0;
38 For j =1 to m do 6.2. Weighting QoS criteria
39 If Si <> Sk then
40 nap = nap + Fj (Si, Sk) * Weight [Qosj]; As described in section 2.2, the first step of the AHP method
41 end for
42 P (Si, Sk) = nap;
is to decompose the problem in hierarchal form. In our case
43 end for the problem of skyline service selection is on the top; the set
44 end for of criteria (response time, throughput, accessibility, good
45 //Calculation of the positive, negative and net flow
46 For i = 1 to n do
practices, material and class) in the intermediary level and the
47 temp1 = 0, temp2 = 0 last level contains the skyline services candidates (S1, S3, S4, S6
48 For j going of 1 in n do and S7).
49 temp1 = temp1 + P (Si, Sj); Table 4 shows the weights of criteria obtained (step 2 &3).
50 temp2 = temp2 + P (Sj, Sj);
51 end for Table 4. Weighting of QoS criteria.
52 PHpositive (Sj) = temp1;
time
Response
throughput
Accessibility
practices
Good
Documentati
Class
Weights
53 PHnegative (Sj) = temp2;
54 PHnet (Sj) = temp1 - temp2;
55 end for
56 Ranking Skyline services according to PHnet.
Response 1 3 2 4 5 3
time 0,35
6. A demonstrative case study
throughput 1/3 1 1/2 3 4 1/2 0,14
In this demonstrative example, we present a case study to
Accessibility 1/2 2 1 4 6 3 0,26
illustrate our approach using QWS DataSet [22].
The Web services of Table 3 give the same functionality to Good 1/4 1/3 1/4 1 2 1/2
consult meteorological data with different QoS metrics values. practices 0,07
They accept in input parameters the city and the country, and Material 1/5 1/4 1/6 1/2 1 1/2 0,05
return as response a set of information elements on the
Class 1/3 2 1/3 2 2 1 0,13
22
As for the test of consistency (step 4 of the AHP method), In this step, the preference degree (Outranking degree) is
it is illustrated by the following calculus: computed between each two Skyline services (See Table 5),
max = (6,37 + 6,20 + 6,35 + 6,15 + 6,19 + 6,41)/6 = 6,28. the expression (3) is used as follows:
To calculate the coherence indicator, the expression (1) is Pi (S6, S7) = Wi * Fi (S6, S7) =0,35*0 + 0,14*0,172 +
used as follows: 0,26*1 + 0,07* 1 + 0,05*0 + 0,13*0 = 0,354.
IC = (max-1) / (N-1) = (6,28-1) / (6-1) = 0,06
While the Coherence ratio is calculated as follows Table 6. Degree of preference calculation.
according to the expression (2):
S1 S3 S4 S6 S7 + = + - -
RC = 100 * (IC / ACI) = 100 * (0,06 / 1,24) = 0,04.
Coherence ratio is equal to 0,04 which is less than 0.1, this S1 0,490 0,750 0,750 0,820 2,810 2,726
means that weights allocated to QoS criteria are consistent and 0,014 0,260 0,610 0,680 1,564 0,615
S3
user judgments are sufficient. The QoS Weights can be used in
the next step to rank Skyline services. S4 0,035 0,392 0,610 0,694 1,739 0,724
time
Response
throughput
Accessibility
practices
Best
Documentation
Class
23
We consider that all services for the QoS DataSet offer the Skyline results, due to the time spent in computing the
same functionality. For the Skyline computing we used BNL preference intensity and the preference degree between each
algorithm, due to its simplicity and large usability [12]. For couple of Skyline services. This execution time is reduced
eight QoS metrics, the skyline result returns 117 services considerably by reducing the decision space focusing only on
among 2507 services, for 3 and 4 metrics, the skyline result interesting services that are not dominated by any other
contains only 12 services among 2507. service. If we apply PROMETHEE directly as it is proposed
in existing approaches [8], [9] and [17] on a complete datasets,
the execution time will increase more and more.
The performance evaluation in Figure 5 shows that our
approach has overcome the computational limitation of
MCDM aggregation methods [24] by applying them
(PROMETHEE) on a reduced number of alternatives using
Skyline method. For example, for eight QoS metrics and 2705
services, the space decision is reduced to 127 Skyline services
only.
24
steals(STL), blocks (Blk), turnover, Free Goals Attempted [8] R. Karim, Ding, C., and Chi, C.H. An enhanced Promethee model for
QoS-based web service selection. InServices Computing (SCC), 2011
(FGA), Free Throws Made (FGM), Free Throws Attempted International IEEE Conference one (pp. 536-543). IEEE. On 2011.
(FTA), Free Throws Made (FTM). The skyline result returns [9] Vesyropoulos, Nikolaos et Georgiadis, Christos K. QoS-Based Filters in
252 players among 21 961. After assigning weights for each Web Service Compositions: Utilizing Multi-Criteria Decision Analysis
dimension using AHP method, the Skyline players are ranked Methods. Journal of Multi-Criteria Decision Analysis, 2014.
using PROMETHEE II. [10] R.W Saaty,. The analytic hierarchy process-what it simple percentage
The ranking result depends on the weights assigned and the and how it used simple percentage. Mathematical Modelling, 9 (3), 161-
176. On 1987.
preferences (Preference and Indifference threshold) expressed
[11] J P. Brans, Vincke Ph. Mareschal B.: «How to select and how to rank
by the user on the 10 dimensions. Comparing the ranking projects: the Promethee methods», European newspaper of operational
results of our approach with other works of [16] and [25], as research,1986.
shows in Table 9, we can see that the most players reported by [12] S. Borzsonyi Orzs Onyi, D. Kossmann, and K. Stock. "The Skyline
this works (G. Mcginnis 1974, M. Jordan 1988, M. operator". IN ICDE. On 2001.
Jordan1987, M. Jordan1989) are returned by our approach. [13] Tan, Kian-Lee, ENG, Pin-Kwang, OOI, Beng Chin, and al. “Efficient
progressive skyline computation”. In: VLDB. P. 301-310. On 2001.
[14] D. Papadias, There. CAT, G. Fu, and B. Seeger. “Optimum year and
8. Conclusion progressive algorithm for Skyline queries”. IN SIGMOD ’03:
Proceedings of the 2003 International ACM SIGMOD Conference one
In this paper, the problem of Skyline services ranking is Management of Dated, pages 467-478, New York, NY, USA. ACM. On
handled, taking into consideration the preferences and the 2003.
subjectivity of the user. Our hybrid proposed approach, called [15] K. Benouaret “Advanced techniques for Web service query
SkyAP-S3, combines three methods of Multi-Criteria Decision optimisation”. Doctoral dissertation. Claude Bernard Lyon 1 university.
Computer science laboratory in Picture and information Systems. On
Making: the Skyline method is used to reduce the number of 2012.
candidate services to be considered in the decision space, the [16] G. Dai, Qingsheng Zhu. “Using Skyline and Dominance Relationship
AHP method for weighting selection criteria and the for Web Ranking services”. Newspaper of Computational Information
PROMETHEE method for the ranking of Skyline services, Systems 9: 10 (3977-3984. On 2013.
exploiting outranking relationships and preferences functions [17] Seo, Young-Jun, Hwa-Young Jeong, and Young-Jae Song. "Best Web
(preference and indifference thresholds). This approach allows service selection based on the decision making between QoS criteria of
service."Embedded Software and Systems. Springer Berlin Heidelberg,
giving the user the best Skylines services, with a better 408-419. 2005.
compromise between different selection criteria. We [18] Chakhar, S. Youcef, V. Mousseau, L. Mokdad and S. Haddad,
developed our ranking algorithm based on PROMETHEE “Multicriteria Evaluation-Based Conceptual Framework for Composite
method. The experimental evaluation illustrates that our Web Service Selection”, http://www.lipn.univ-paris13.fr/
youcef/BookQoS.2011.
approach can capture the user preferences and retrieves the
[19] Seo, Young-Jun, Hwa-Young Jeong, and Young-Jae Song. "A study on
best Skyline services. As for a future work, we plan to extend web services selection method based on the negotiation through quality
this work to QoS-based Web service composition; another line broker: A maut-based approach." Embedded Software and Systems.
of future research is to perform large-scale experiments with a Springer Berlin Heidelberg, 2005. 65-73.
large number of Web services. [20] C. Marchairs, J. Springael, K. From Brucker, In. Verbeke, “Promethee
and AHP: The Design of Operational Synergies in Multicriteria
Analysis. strengthening Promethee with ideas of AHP”, European
9. References Journal of Operational Research, Flight. 153, Exit 2, pp.307-317, 2004.
[1] F. Curbera, Mr Duftler, highway. Khalaf, W. Nagy, N. Mukhi, and S. [21] Wang, Jian-Jun et Yang, De-Li. “Using a hybrid multi-criteria decision
Weerawarana, “Unraveling the Web Web services - A Year introduction aid method for information systems outsourcing”. Computers &
to SOAP, WSDL, and UDDI”, Internet IEEE COMPUTING, flight. 6, Operations Research, 2007, vol. 34, no 12, p. 3691-3700.
N. 2, Mar.-Apr. pp. 86-93. On 2002 [22] E. Al-Masri, Mahmoud, Q. H., "QoS-based Discovery and Ranking of
[2] M. Alrifai, D. Skoutas, and T. Should laugh, “Selecting skyline services Web Services", IEEE 16th International Conference one Computer
for QoS-based web service composition, ”in Proceedings of the 19th Communications and Networks (ICCCN), pp. 529-534. (for QWS
international conference one World wide web, pp. 11-20, 2010. DATASET VERSION 1.0 and QWS DATASET VERSION 2.0), on
2007.
[3] Q. Yu and A. Bouguettaya, “Computing service skyline from uncertain
qows,” IEEE T. Computing services, flight. 3, N. 1, pp. 16-29, 2010. [23] Al-Masri, E., Mahmoud, Q.H. "Crawling Multiple Uddi Business
Registries". In: Procs of the 16th International Conference one World
[4] L. Shao, J. Zhang, There. Wei, J. Zhao, B. Xie, and H. Mei. Wide Web (WWW2007), May 2007, Banff, Alberta, Canada. ACM,
“Personalized QoS prediction forweb services via collaborative pp.1255-1256, on 2007.
filtering”. In ICWS, pages 439-446, on 2007.
[24] MARINONI, Oswald. A discussion on the computational limitations of
[5] K. Benouaret, Benslimane, Djamal, and Hadjali, Allel. “Selecting outranking methods for land use suitability assessment. International
Skyline Web Services for Multiple Users Preferences”. In: Web Services Journal of Geographical Information Science, 2006, vol. 20, no 1, p. 69-
(ICWS), 2012 IEEE 19th International Conference one. IEEE,p. 635- 87.
636. On 2012.
[25] Vlachou, Akrivi, and Michalis Vazirgiannis. "Ranking the sky:
[6] Q. Yu and Athman Bouguettaya, “Efficient Foundations for Web Discovering the importance of skyline points through subspace
Selection service ”. Springer, 2010. dominance relationships." Data & Knowledge Engineering 69.9 (2010):
[7] Godse, Manish, Rajendra Sonar, and Shrikant Mulik. "The analytical 943-964.
hierarchy process approach for prioritizing features in the selection of
web service." on Web Services, 2008. ECOWS'08. IEEE Sixth European
Conference. IEEE, 2008.
25