Open Framework for Web Service Selection Using Multimodal and Configurable Techniques
Oscar Cabrera 1, Marc Oriol 1, Xavier Franch 1, Jordi Marco 1, Lidia López 1, Olivia Graciela Fragoso Díaz 2, and René Santaolaya 2 1 Universitat Politècnica de Catalunya (UPC), Barcelona, Spain
2 Centro Nacional de Investigación y Desarrollo Tecnológico (CENIDET), Morelos, Mexico {ocabrera, moriol, franch, llopez}@essi.upc.edu, [email protected], {ofragoso, rene}@cenidet.edu.mx
Abstract. Services as part of our daily life represent an (QoS), non-functional requirement (NFR), service level important means to deliver value to their consumers agreement (SLA), ranking services. and have a great economic impact for organizations. The service consumption and their exponential proliferation show the importance and acceptance by 1 Introduction their customers. In this sense, it is possible to predict that the infrastructure of future cities will be supported In today´s world, there are different kinds of by different kind of services, such as smart city services created to facilitate the life of citizens in services, open data services, as well as common their daily tasks. These services have been services (e.g., e-mail services), etc. Nowadays a large developed to solve different needs according to percentage of services are provided on the web and certain requirements of different human desires. are commonly called web services (WSs). This kind of As a result, an enormous explosion in offering services has become one of the most used services has occurred. In fact, it can be observed technologies in software systems. Among the that for a given need, a plethora of services can challenges when integrating web services in a given system, requirements-driven selection occupies a be found. In addition, according to [1] there is a prominent place. A comprehensive selection process growth in consumer services driven by various needs to check compliance of Non-Functional social, economic, and technological factors (e.g., Requirements (NFRs) which can be assessed by a demand for social services, the size and role of analyzing the Quality of Service (QoS). In this paper, the public sector, complexity of work we describe a framework called WeSSQoS that aims at environments, etc.). ranking available WSs based on the comparison of A generic definition of a service is provided by their QoS and the stated NFRs. The framework is the Office of Government Commerce (OGC) in its designed as an open Service Oriented Architecture ITIL standard as follows [1]: “A service is a means (SOA) that hosts a configurable portfolio of normalization procedures and ranking algorithms which of delivering value to customers by facilitating can be selected by users when starting a selection outcomes customers want to achieve without the process. The QoS data from WSs can be obtained ownership of specific costs and risks” . either from a static, WSDL-like description or The OGC considers that the outcomes dynamically through monitoring techniques. WeSSQoS mentioned are possible through the performance is designed to work over multiple WS repositories and of different tasks and are limited by the presence QoS sources. The impact of having a portfolio of of certain constraints. In this sense, the presented different normalization and ranking algorithms is paper is focused on quality constraints that illustrated with an example. characterize services. Specifically, web services
Keywords. Web service (WS), web service selection, (WSs), since a large amount of services are being service oriented architecture (SOA), quality of service provided using this technology.
Computación y Sistemas, Vol. 18, No. 4, 2014, pp. 665–682 ISSN 1405-5546 doi: 10.13053/CyS-18-4-2057 666 Oscar Cabrera, Marc Oriol, Xavier Franch, Jordi Marco, Lidia López, Olivia Graciela Fragoso Díaz, and René Santaolaya
WSs integrate a set of protocols and standards The main goals that we aim to address in this for data interchange among software applications paper are: how NFRs are expressed; what developed in different programming environments measure of the satisfaction of an NFR in a given and languages, and executed in different WS is; how these individual measures are platforms. This interoperability is provided mainly combined in order to rank the WS according to a by the following open standards: XML, SOAP, set of NFRs; how the QoS of a WS may be HTTP, WSDL, and other web-related obtained; where the WSs are obtained from; and standards [2]. what the value obtained by combining different normalization procedures and ranking algorithms WSs have become a useful technology to to select WSs is. To attain these goals we have implement any kind of software, providing greater designed WeSSQoS (Web Service Selection interoperability and scalability. This success has based on Quality of Service), a framework for triggered the emergence of a huge WSs selecting WSs based on their QoS and NFRs. marketplace. Consequently, for a given WeSSQoS proposes an open Service Oriented functionality we may find a large set of WSs that Architecture (SOA) that is able to manage several can be selected in several ways. This proliferation ranking algorithms and normalization procedures of WS increments the chances to find existing for computing the adequacy of a WS with respect software that satisfies the stated needs, but at the to NFRs. These NFRs are expressed by means of same time raises new problems and challenges. formulae stated over QoS attributes (i.e., SLA Among them, there is an increasing need for clauses) coming from the quality model proposed selecting the most appropriate WS in a given in an earlier work [5]. NFR are classified as context of usage [3]. Usually this problem is mandatory and optional, and this information may studied in relation with the requirements elicited be used for ranking the results. WeSSQoS is from the stakeholders. In other words, the goal is designed to work over several WS repositories to select the WS that “better” satisfies the that eventually can be built using different stakeholder requirements. technologies. In order to get the behavior of the We consider here the classical distinction accessible WSs with respect to the selection among functional and non-functional criteria, it is possible to use either the description requirements [4]. With respect to functional of the QoS (if included in the WS definition), or requirements, it is necessary to validate that a the results of WS monitoring (obtaining then the WS fulfills the functionality expected by the real, updated QoS of the WS). In this sense, we stakeholders. On the other hand, non-functional share the vision of [6] that proposes to define a requirements (NFRs) refer to the Quality of priori only static attributes like cost, whilst Service (QoS) that a given WS offers, i.e., dynamic attributes like response time or behavioral and non-behavioral characteristics that availability should be obtained through the WS exhibits for offering a given functionality: monitoring. cost, response time, availability, etc. Usually, The rest of the paper is organized as follows. NFRs are expressed in terms of conditions over In Section 2 a review of some similar frameworks the QoS in a document named Service Level is provided. Then, Section 3 describes the Agreement (SLA). Therefore, we can assess if a proposed WS selection process. Section 4 WS w satisfies an NFR r by checking if the QoS introduces normalization procedures and ranking of w satisfies the clauses from the SLA that refer algorithms. Section 5 describes the framework to the concepts inherent to r. architecture proposed. Sections 6 and 7 describe Given this context, our work proposes a a prototype and provide some validation. Finally, framework for ranking a set of WSs that belong to Section 8 presents conclusions and future work. a certain software domain. We assume that the functional requirements are used to determine 2 Related Work this software domain, therefore our framework focuses on the ranking based on the satisfaction In the academic research, different frameworks of NFRs. for ranking and selecting WSs according to their
Computación y Sistemas, Vol. 18, No. 4, 2014, pp. 665–682 ISSN 1405-5546 doi: 10.13053/CyS-18-4-2057 Open Framework for Web Service Selection Using Multimodal and Configurable Techniques 667
QoS have been proposed. In Table 1 we show a b. Attributes: quality attributes considered in representative sample of these proposals, those systems. In some cases, a small including our WeSSQoS framework, comparing predefined set of quality attributes is used, them according to the following criteria: whereas other frameworks allow the usage of arbitrary ones (although they may validate the a. Architectural style: architecture in which proposal with a given set). We represent the the framework has been developed. We find value of this criterion by using the amount of Component Based Architectures (CBA), attributes defined as dynamic (d) or as static Service Oriented Architectures (SOA) and a (s). In case of configurable attributes, i.e., the combination of both. We represented each of possibility of adding new attributes, we use an the styles by using C, S, and CS, respectively. asterisk (*). It is worth noting that adopting SOA allows c. QoS data source specifies if quality data integrating heterogeneous systems more are declared in the service description easily. (represented by using S) or for dynamic quality
Table 1. Comparative table of frameworks
Proposal (a) (b) (c) (d) (e) (f) (g)
E. Al-Masri et al. [6] C 6d 3s SM X X V V
T. Yu et al. [7, 8] C 4d 1s S X V X X
X. Wang et al. [9] - *1d5s S X X X X
D. D’ Mello et al. [10] CS *3d2s SM X X X X
H. Wang et al. [11] C 6s SM X X X X
P. Wang et al. [12] - *6d S X X X X
R. Mohanty et al. [13] - 9s S X V X X
Q. Tao et al. [14] C 6s SM X X X X
H. Cai et al. [15] CS *0s SM X X X X
L. Sha et al. [16] CS 7s SM X X X X
M. Alrifai et al. [17] C 0s SM X X X X
A. Huang et al. [18] - 0s S X X X X
C. Lin et al [19] C 0s S X X X X
Z. Gao et al. [20] - 5s S X X X X
WeSSQoS CS *9d1s SM * V * V *V V
Computación y Sistemas, Vol. 18, No. 4, 2014, pp. 665–682 ISSN 1405-5546 doi: 10.13053/CyS-18-4-2057 668 Oscar Cabrera, Marc Oriol, Xavier Franch, Jordi Marco, Lidia López, Olivia Graciela Fragoso Díaz, and René Santaolaya
attributes, if their value is obtained through not a SOA, it does not allow adding external monitoring (represented by using M). In cases algorithms in an easy manner. where the proposals provide both kinds of Regarding the criterion of prototype available , sources we represent the value by using SM. we identified that although in most of the d. Multinormprocedure specifies if the proposals a prototype is being described and framework is able to work with more than one even some of them have a web page (e.g., [7, 8]), normalization procedure in order to obtain the there is not a framework available. In fact, we normalized QoS data about WSs and have only found a tool available from Al-Masri et stakeholders. We represent the value of this al . [6]. criterion by using yes (V) or no (X). In case of proposals which allow adding new procedures, 3 The Proposed Web Service Selection we use an asterisk (*). Process e. Multialgorithm specifies if the framework is able to work with more than one selection Figure 1 shows the proposed web service algorithm. We represent the value of this selection process with the following inputs and criterion by using yes (V) or no (X). In case of selection phases: proposals which allow adding new algorithms, we use an asterisk (*). Inputs:
f. Multirepository specifies if the framework - WSlist, a QoS matrix of size k×n , where (w 1, is able to obtain data from different …, w k) are the candidate WS and (q 1, …, q n) repositories and combine information to extend are the quality attributes referred in the NFRs. the number of services and quality attributes to WSlist[ i, j ] stands for the value of the quality evaluate. We represent the value of this attribute q j in the WS w i. criterion by using yes (V) or no (X). In case of - lreqs, a NFR vector of size n, where lreqs[ i] proposals which allow adding new specifies (1) the value that is required for the repositories, we use an asterisk (*). attribute q i, (2) a Boolean value that indicates g. Prototype available specifies if the if the attribute’s value is to be minimized or framework is available to be used for the maximized, and (3) another Boolean value research community. We represent the value that indicates if the required attribute’s value of this criterion by using yes (V) or no (X). is mandatory or not. A value is mandatory when it cannot be higher than the required As a result of the previous evaluation we threshold when it has to be minimized, or identified different gaps, such as a lack of cannot be lower than the threshold when it frameworks with the capability to retrieve a list of has to be maximized, e.g., a NFR may state web services from different sources. As far as we to minimize the cost mandatorily with a know, the only framework that fulfills this criterion maximum of 100 euros per month. is provided by Al-Masri et al . [6], whose framework obtains a list of WSs from several Selection phases: sources (UDDIs, ebXMLs, search engines, and - Normalization. This phase has the purpose of service portals). However, it does not specify the integrating the heterogeneous QoS attributes’ method to combine the services data when values over which decision-making in the different sources have the same service with WSs selection problem relies. Both inputs different QoS data: cost, brand reputation, etc. WSlist and lreqs must be normalized to Another important gap is a lack of frameworks compensate the different measurement units with the capability to reuse existing normalization of the different QoS values by projecting them procedures and selection algorithms, which would onto a normalized interval. Interval allow assessing results obtained from different boundaries are established by the proposals. In fact, only QCWS [7, 8] offers the normalization procedure used. Details of the capability of multialgorithm . However, since it is different normalization procedures are
Computación y Sistemas, Vol. 18, No. 4, 2014, pp. 665–682 ISSN 1405-5546 doi: 10.13053/CyS-18-4-2057 Open Framework for Web Service Selection Using Multimodal and Configurable Techniques 669
Fig. 1. Flow diagram for the web services selection process
described in the next section. The result of 4 Normalization Procedures and this phase is the normalized structures Ranking Algorithms denoted by WSlistN and lreqsN. - Ranking. Starting from the normalized data in One of the main characteristics of the proposed the previous stage, a ranking algorithm can framework architecture is that it supports the be applied with the goal of computing a coexistence of normalization procedures and similarity measure between the NFR (lreqsN) ranking algorithms offered as services. and the QoS of each service (WSlistN). This algorithm may be any of the commonly 4.1 Normalization Procedures employed in Vector Space Models (VSM) to evaluate the similarity between two objects The normalization service that WeSSQoS described by vectors [21]. For example, the currently offers has four normalization procedures Euclidian Distance algorithm looks for the (see equations 1 to 4). Nevertheless, as shortest distances between the QoS of each mentioned before, users can extend it by candidate WS and the user NFR. As a result, providing their own normalization procedures. In we obtain the values of the algorithm and the this sense, providing or selecting these WSs ranked according to them. Next section procedures depend on both the user’s needs and describes different ranking algorithms. the properties of such procedures, i.e., the - Priority evaluation . In this phase two main selection process involves analyzing and types of WSs ranking are carried out, one by evaluating their advantages and disadvantages the number of mandatory requirements that as well as their applicability. services fulfill and one by the selection 0< ≤1, (1) algorithm used. ⁄
Computación y Sistemas, Vol. 18, No. 4, 2014, pp. 665–682 ISSN 1405-5546 doi: 10.13053/CyS-18-4-2057 670 Oscar Cabrera, Marc Oriol, Xavier Franch, Jordi Marco, Lidia López, Olivia Graciela Fragoso Díaz, and René Santaolaya
0< ≤1, (1a) , ⁄ ∑ ( ∗ ) (5) ( , ) ∑ ∗∑ 0≤ ≤1, (2) − ⁄ −
0≤ ≤1, (2a) ∑ ( ∗ ) (6) − ⁄ − ( , ) , ∑ , ∑
0< <1, (3) ∗∑ ( ∗ ) (7) ( , ) ∑ ∑ ,
0< <1. (4) (8) ∑ ( ∗ ) ( , ) , ∑ + ∑ − ∑ ( ∗ ) For example, according to [22], procedure (1) is very common and has an intuitive (9) interpretation. It also maintains the proportionality ( , ) ( − ) , of different values, i.e., , for all i, k. Procedure (2) refines the / previous / one in order that the normalized scale covers exactly the . interval [0, 1], i.e., for each criterion the worse (10) ( , ) value is 0 and the best value is 1, but in this case ∑ ( ) the proportionality is not maintained. Procedure (3) offers almost the same advantages as According to [24], the cosine measure (5) procedure (1), although (3) concentrates assumes that similarity is proportional to the angle towards small values. Finally, procedure (4) offers between two t-dimensional vectors in a t- an important advantage allowing dimensionless dimensional space. Because the numerator is comparisons of vectors related to the problem divided by the product of the lengths of the criteria. Procedures (1a, 2a) represent the case of vectors, the measure tends to give low similarities minimum values of procedures (1, 2), between long vectors, i.e., vectors with many respectively, i.e., they vary the relation mentioned terms. The overlap measure (6) compensates the above and establish 1 as the worst value and 0 as cosine measure by dividing by the vector having the best value. An extended comparative analysis the lowest sum of weights. The Dice coefficient of these procedures is out of the scope of the (7) gives more weight to matches in the data than paper, but the reader can refer to [22, 23] for to differences, whereas Jaccard's coefficient (8) is details regarding the different normalization the proportion of characters (i.e., index terms) that procedures. match, excluding those characters that lack in both vectors. Finally, the Euclidean distance (9) emphasizes differences between two vectors 4.2 Ranking Algorithms more than matched features. An important disadvantage of this measure is related to the The ranking service which WeSSQoS currently variables used, i.e., if these variables are offers includes six ranking algorithms (see correlated then the information provided will be equations 5 to 10). As mentioned before, users redundant. A variation of the Euclidean distance, can also provide their own ranking algorithms. In emphasizing distance rather than similarity is this sense, users are responsible for selecting the presented in equation (10). An extended ranking algorithms fulfilling their requirements, by comparative analysis of these algorithms is out of analyzing and assessing the advantages and the scope of the paper. Details of such algorithms disadvantages as well as their applicability. can be checked in [24, 25].
Computación y Sistemas, Vol. 18, No. 4, 2014, pp. 665–682 ISSN 1405-5546 doi: 10.13053/CyS-18-4-2057 Open Framework for Web Service Selection Using Multimodal and Configurable Techniques 671
Table 2. Inputs and outputs of the normalization phase
NFRs from Stakeholder, lreqs [30, 35, 31, 15, 20, 0.5, 0.03, 150] QoS from candidate WSs, WSlist
WS1 20 30 25 15 10 0.4 0.3 50 WS2 5 10 20 20 15 0.5 0.2 80
WS3 33 11 6 8 10 0.8 0.4 125 Fig. 2. WeSSQoS general architecture WS4 25 35 45 45 15 0.5 0.5 302
To illustrate the execution of the WS selection process, let’s consider the following example. A Normalized NFRs, lreqsN user needs to select a WS for a given domain [0.91, 1, 0.69, 0.33, 1, 0.62, 0.60, 0.50] with a set of NFRs instrumented in different metrics (e.g., cost, response time, availability, Normalized QoS from candidate WSs, WSlistN etc.). The user defines the list of values for such metrics in Ireqs (see Table 2, NFRs from WS1 .7 .9 .6 .3 .5 .5 .6 .2 Stakeholders Ireqs ). WS2 .1 .3 .4 .4 .8 .6 .4 .3
In the repository there are 4 WSs that fulfill the WS3 1 .3 .1 .2 .5 1 .8 .4 functionality required by the user with different WS4 .8 1 1 1 .8 .6 1 1
QoS (see Table 2, QoS from candidate WSs). Both NFRs and QoS are normalized by applying the procedure that better fulfills the user’s needs. Table 3. Results using ranking and priority In Table 2 we show the results applying the evaluation normalization procedure (1). These normalized data are then the input for the ranking phase (see ID Name Euclidian Ordering distance by QoS Table 3). On top of the table, we depict the results WS of applying the Euclidean distance algorithm (9). It WS1 AirportWeather 0.71083 1 may be observed that WS1 has the minimum Check value, thus it looks like a promising candidate for selection before evaluating the compliance WS2 BerreWeather 1.14562 4 degree of the mandatory requirements. WS3 FastWeather2 1.11749 3
As for the priority evaluation phase, let’s WS4 Weather 1.01981 2 suppose that all requirements are mandatory. ID Name Mandatory QoS vs. Based on this premise, the results depicted in the WS Mandat. bottom of Table 3 shows that WS1 and WS2 comply with 5 of the 8 NFR, whilst WS3 and WS4 WS1 AirportWeather 5/8 1 comply with 3. When the results obtained by the Check phases of ranking and priority evaluation are WS2 BerreWeather 5/8 2 combined, the prioritized list of services is as shown at the bottom of Table 3. WS1 is still the WS3 FastWeather2 3/8 4 best ranked service, although the ranking results WS4 Weather 3/8 3 for the rest of services change.
Computación y Sistemas, Vol. 18, No. 4, 2014, pp. 665–682 ISSN 1405-5546 doi: 10.13053/CyS-18-4-2057 672 Oscar Cabrera, Marc Oriol, Xavier Franch, Jordi Marco, Lidia López, Olivia Graciela Fragoso Díaz, and René Santaolaya
5 Proposed Framework Architecture limited to, six ranking algorithms. Users can provide and add their own ranking algorithms The proposed framework, Web Service Selection which will be available for the scientific Based on Quality of Service ( WeSSQoS ) is community. structured under the SOA paradigm in order to Figure 2 also shows the relationships among facilitate its integration into other systems. the services previously mentioned. As shown, the Figure 2 shows the elements integrating the composition of services follows an orchestration framework architecture which are described as managed by QoSSelector service. A sequence follows: diagram of such orchestration is shown in - QoSSelector is a service that integrates three Figure 3. The main method in QoSSelector is services: QoSRepositoryProxy , QoS rank4QoSRepository which is used to rank the NormalizeData , and QoSSelectionModel , services . The input of this operation is a list of providing a unified view and a single entry repositories (lProxies ), the list of requirements point to the whole system. (lReqs ), domain of the WSs ( domain ), - QoSRepositoryProxy is a service that obtains normalization procedure ( iNumNormalize ), and the QoS of WSs that belong to a given ranking algorithm ( iNumUtilFunction ). The output domain. Two sources of QoS information are obtained is a list of WSs ranked according to the defined: satisfaction of NFRs and mandatory nature, - Monitor. It obtains the QoS at execution according to the process described in Section 3. time by means of monitoring techniques. The sequence shown in Figure 3 is described A monitor works on a predefined catalog as follows: rank4QoSRepository operation of dynamic quality attributes. Any invokes getServicesDataFromDomain operation information about static quality attributes for each QoSRepositoryProxy (Databank or will be available in the description of the Monitor ) specified in lProxies . From such service, e.g., service cost. invocation, the list of services with QoS - Data Bank. It obtains the QoS from the information is obtained ( WSlist ). In case of having WS provider which describes quality data repeated QoS information in more than one in extended WSDL files. In case of repository, a simple priority policy is applied to the dynamic quality attributes, such as mean repositories list, i.e., the order in the repositories response time , the quality value is the list determines the priority of attributes appearing one that the provider promises to deliver. in more than one repository. - QoSNormalizeData is a service that Once the list WSlist of services with their QoS normalizes stakeholder requirements and information is obtained, the operations of QoS data obtained from WSs by applying normalization and ranking are applied. First, the normalization procedures as described in operation getNormalizedData from QoS Section 4. Its SOA is flexible enough as to NormalizedData service is executed. This extend the portfolio of normalization operation takes as input the following parameters: procedures. In its current version, WeSSQoS WSlist , NFRs from the stakeholder represented provides, but is not limited to, four by lreqs , and the type of normalization process normalization algorithms. Users can provide represented by iNumNormalize . The output of this and add their own normalization procedures method is the normalized list of QoS and NFR. which will be available for the scientific Afterwards, QoSSelectionAlgorithm operation community. from QoSSelectionModel service is executed in - QoSSelectionModel is a service that sorts order to rank WSs applying the ranking algorithm candidate WSs by applying ranking identified by iNumUtilFunction . algorithms as described in Section 4. Also, its The final output is a list of orderedWS that can internal architecture is flexible enough as to be simple or multiple. A simple list provides WSs extend the portfolio of ranking algorithms. sorted by a single ranking algorithm using a single Currently, WeSSQoS provides, but is not normalization procedure and furthermore provides
Computación y Sistemas, Vol. 18, No. 4, 2014, pp. 665–682 ISSN 1405-5546 doi: 10.13053/CyS-18-4-2057 Open Framework for Web Service Selection Using Multimodal and Configurable Techniques 673
Table 4. Interfaces of WeSSQoS services Table 4 shows the interfaces of services appearing in the sequence diagram, whereas QoSSelector attributes and classes involved are represented in Figure 4a and 4b. A general description of these Operation: Result: elements is provided as follows: rank4QoSRepository orderedWS: list - lproxies is a list of repositories from which the Input parameters:
Computación y Sistemas, Vol. 18, No. 4, 2014, pp. 665–682 ISSN 1405-5546 doi: 10.13053/CyS-18-4-2057 674 Oscar Cabrera, Marc Oriol, Xavier Franch, Jordi Marco, Lidia López, Olivia Graciela Fragoso Díaz, and René Santaolaya
Fig. 3. Sequence diagram of the basic use case of the framework internal repositories (i.e., local to WeSSQoS ) and interval. The framework allows using both internal external ones (i.e., provided by stakeholders). normalization procedures (i.e., local to As already mentioned, the repositories are WeSSQoS ) and external ones (i.e., provided by identified using their endpoint. Each repository the stakeholders). might have different strategies to extract QoS Normalization procedures are also identified data, i.e., using the strategy design pattern it is using their endpoint. Each procedure selected or possible to extend the repository behavior provided might have optional strategies acting as adopting different QoS data sources in the same a repository of normalization procedures in the repository (e.g., QoS data from XML documents, same endpoint. databases, etc.). Finally, each repository from the The third section, depicted in Figure 7, list of chosen repositories can be prioritized. corresponds to ranking algorithms that will be The second section, depicted in Figure 6, applied to prioritize WSs. The basic use case of corresponds to normalization procedures that will this section is to provide at least a ranking be applied on both QoS data from WS and NFR algorithm fulfilling the data structure specified in from stakeholders. The basic use case of this the QoSSelectionModel service depicted in section is to provide at least a normalization Table 4. procedure in order to compensate the different Furthermore, the framework allows using both measurement units of the different QoS and NFR internal ranking algorithms and external ones. values by projecting them onto a normalized Ranking algorithms are identified using their
Computación y Sistemas, Vol. 18, No. 4, 2014, pp. 665–682 ISSN 1405-5546 doi: 10.13053/CyS-18-4-2057 Open Framework for Web Service Selection Using Multimodal and Configurable Techniques 675
Fig. 4a. Class diagram of the services supporting the internal architecture of WeSSQoS endpoint. Each algorithm selected or provided them and the QoS information from WSs. For might have alternative strategies acting as a each attribute introduced, the following repository of ranking algorithms in the same information is required: the value that WSs should endpoint. meet, maximization or minimization to Finally, each endpoint from the list of chosen compensate the attribute value, and information selection models can be deleted. that allows identifying when an attribute is mandatory to prioritize services. The fourth section, depicted in Figure 8, corresponds to stakeholder requirements , where Finally, the results section depicted in Figure 9 stakeholders introduce NFRs to be fulfilled. These shows the resulting ranking and provide different NFRs are settled over quality attributes that can options described below. The first ranking be attributes provided by the framework based on provided is a sorted WSs list according to the [5] or by other external source. Clearly, ranking algorithm and normalization procedure stakeholders have the responsibility of choosing chosen by stakeholders. Also the number of quality attributes that WSs should comply with, mandatory attributes fulfilled is depicted. i.e., these attributes will be used to compute the Figure 10 shows the graphic option of the relationship (similarity or dissimilarity) between results with two types of charts. The chart on the
Computación y Sistemas, Vol. 18, No. 4, 2014, pp. 665–682 ISSN 1405-5546 doi: 10.13053/CyS-18-4-2057 676 Oscar Cabrera, Marc Oriol, Xavier Franch, Jordi Marco, Lidia López, Olivia Graciela Fragoso Díaz, and René Santaolaya
Fig. 4b. (cont.) Class diagram of the services supporting the internal architecture of WeSSQoS right shows the ranking results applying normalization procedure (1) and ranking algorithm 7 Validation (5). The chart on the left shows the ranking results applying mandatory requirements. In order to test our prototype, we designed a scenario to execute some test cases. The As mentioned before, the architecture of scenario was designed to assess the following WeSSQoS allows providing both a list of features of our framework: normalization procedures and a list of ranking algorithms supplying the list of results. This - Quality attributes management . In the functionality allows comparing the different scenario, the customer can decide the quality rankings obtained and the behavior shown by a attributes which she is interested in. These particular ranking algorithm in combination with a attributes may or may not be defined in the normalization procedure. In this sense, Figure 11 information about the WSs being selected. shows the ranking of four services applying two The basic case is when the customer asks for ranking algorithms with two normalization a subset of attributes defined on the procedures yielding four different results. repositories. The customer can also ask for
Computación y Sistemas, Vol. 18, No. 4, 2014, pp. 665–682 ISSN 1405-5546 doi: 10.13053/CyS-18-4-2057 Open Framework for Web Service Selection Using Multimodal and Configurable Techniques 677
Fig. 5. Repositories of web services with QoS description
Fig. 6. Normalization procedures interface
attributes that are not specified on - The WS of each repository can be repositories, these attributes will be treated as different. In this case we consider as WS undefined by the ranking algorithm. candidates the union of all services inside all repositories. - Repositories independence. Our framework - More than one repository may contain does not have restriction on the number of information of a given WS, but the quality repositories used for the search. Each attributes are disjoint. In this case, the repository can be static or dynamic. When algorithm will simply combine the required there is more than one repository, the attributes retrieving them from the following assumptions are considered: adequate repositories.
Computación y Sistemas, Vol. 18, No. 4, 2014, pp. 665–682 ISSN 1405-5546 doi: 10.13053/CyS-18-4-2057 678 Oscar Cabrera, Marc Oriol, Xavier Franch, Jordi Marco, Lidia López, Olivia Graciela Fragoso Díaz, and René Santaolaya
Fig. 7. Ranking algorithms interface
Fig. 8. Stakeholder requirements interface
Fig. 9. Interface for representation of results
- More than one repository may contain Figure 12 shows the architecture implemented information of a given WS, and some and the necessary data for running the tests quality attribute may appear in more than previously described. We have both types of one repository. In this situation, the value QoSRepositoryProxy (static and dynamic). The is taken from the repository with a higher Monitor instances use Axis, whilst the DataBank priority (i.e., the one declared first). (which contains information about two WS
Computación y Sistemas, Vol. 18, No. 4, 2014, pp. 665–682 ISSN 1405-5546 doi: 10.13053/CyS-18-4-2057 Open Framework for Web Service Selection Using Multimodal and Configurable Techniques 679
Fig. 10. Ranking results using Euclidean distance
Fig. 11. Ranking results considering mandatory requirements domains) uses Glassfish. In Figure 12, the names If the priority of repositories (i.e., their order of of some of WSs were included. These services appearance) is Monitor1, Monitor2, DataBank1, were selected in order to highlight services given the service AirportWeatherCheck (which is located in more than one repository, and some of located in all the repositories), ART, CRT, and CA them have attributes in more than one repository. will be taken from the Monitor1, and the other Databank1 contains information about all attributes, from the DataBank1. attributes with the exception of Current ResponseTime (CRT) and CurrentAvailability However, if the order was Monitor2, Monitor1, (CA). In the services from Monitor1 and Monitor2, and DataBank1, the CRT would be taken from the the information about what attributes have Monitor2, ART and CA, from the Monitor1, and information is included too. In addition to the CRT the rest, from the DataBank1. The users can test and CA, there is also information about the the scenarios described before or test other ones AverageResponseTime (ART) in some services. using the WeSSQoS prototype.
Computación y Sistemas, Vol. 18, No. 4, 2014, pp. 665–682 ISSN 1405-5546 doi: 10.13053/CyS-18-4-2057 680 Oscar Cabrera, Marc Oriol, Xavier Franch, Jordi Marco, Lidia López, Olivia Graciela Fragoso Díaz, and René Santaolaya
Fig. 12. Scenario for WeSSQoS tests
service definition (WSDL) or by monitoring systems. The usage of a common interface 8 Conclusions and Future Work (proxy) to retrieve data in a uniform way from these sources provides extensibility to add new In this paper we presented WeSSQoS , a kinds of repositories, independently of the framework for ranking available WSs through the approach used to obtain the data. evaluation of their QoS with respect to the stated - Multinormprocedure. WeSSQoS is able to work NFRs. In terms of the criteria introduced in with any kind of normalization procedure that is Section 2, we can conclude that the proposal has implemented using the defined interface. the following advantages: Eventually, we could use arbitrarily complex - Architectural style. WeSSQoS is developed as procedures, e.g., aggregators of results a Service Oriented System itself. Following through choreography of other WSs defining SOA principles, users can add new services different normalization procedures. related to ranking algorithms, repositories and - Multialgorithm. WeSSQoS is able to work with normalization procedures, if they are compliant any kind of ranking algorithm that is with the expected service definitions. implemented using the defined interface. - Quality attributes. WeSSQoS is independent of Eventually, we could use arbitrarily complex the Quality Model or ontology used to define algorithm, e.g., aggregators of results through quality attributes. The system interface allows choreography of other WSs that define users to select from a well-known predefined different algorithms. set of attributes based on [5], and also add any - Multirepository. WeSSQoS allows the user to kind of quality attributes from any quality include several repositories of WSs with model. As many frameworks, WeSSQoS is independence of the technology used. able to work with either static or dynamic Furthermore, it provides a mechanism to quality attributes, although it’s important to combine the QoS data when the same service mention that this distinction is implicit in the is present in more than one repository. way the data are retrieved. Currently, the user is responsible for selecting - QoS data. WeSSQoS is able to retrieve quality those repositories that are compatible with attributes from either quality descriptions in each other, e.g., repositories should use a
Computación y Sistemas, Vol. 18, No. 4, 2014, pp. 665–682 ISSN 1405-5546 doi: 10.13053/CyS-18-4-2057 Open Framework for Web Service Selection Using Multimodal and Configurable Techniques 681
common terminology to refer to the same 5. Ameller, D. & Franch, X. (2008). Service level quality attribute. agreement monitor (SALMon). Seventh - Prototype available. WeSSQoS is available at International Conference on Composition-Based http://gessi.lsi.upc.edu/wessqos/. The current Software Systems, ICCBSS’08, IEEE, pp. 224– 227. version has been tested and validated as explained in Section 7. 6. Al-Masri, E. & Mahmoud, Q.H. (2009). A broker for universal access to web services. Seventh In Section 5 we dealt with the issue concerning Annual Research Conference on Communication WS repositories’ priority policy, the main idea of Networks and Services, CNSR'09, IEEE, pp. 118– this is to integrate in a general repository the WSs 125. coming from all chosen repositories in a 7. Yu, T. & Lin, K.J. (2005). Service selection prioritized way. It is worth noting that WS algorithms for Web services with end-to-end QoS integration is used in repositories combination constraints. Information Systems and E-Business and it is not part of WS composition, this topic is Management, Vol. 3, No. 2, pp. 103–126. out of the scope of the paper. 8. Yu, T. & Lin, K.J. (2005). A broker-based As future work, we identified several research framework for QoS-aware web service composition. IEEE International Conference on E- lines and improvements that could be performed Technology, e-Commerce and e-Service, EEE'05, in order to increase the current framework’s IEEE, pp. 22–29. capabilities: 9. Wang, X., Vitvar, T., Kerrigan, M., & Toma, I. - Perform tests in large web service (2006). A QoS-aware selection model for semantic ecosystems to ensure the correctness and web services. Service-Oriented Computing suitability of the framework to rank web (ICSOC 2006), Springer , pp. 390–401. services in real situations. 10. D'Mello, D.A., Ananthanarayana, V.S., & Santhi, - Increase the number of dynamic quality T. (2008). A QoS broker based architecture for attributes retrieved by the monitoring system. dynamic web service selection. Second Asia International Conference on Modeling & - Design different sophisticated mechanisms to Simulation, AICMS’08, IEEE, pp. 101–106. combine data from several repositories and 11. Wang, H.C., Lee, C.S., & Ho, T.H. (2007). unify these strategies under a common Combining subjective and objective QoS factors for interface in order to build it as a service. personalized web service selection. Expert - Automate analysis and evaluation of ranking Systems with Applications , Vol. 32, No. 2, pp. 571– algorithms and normalization procedures. 584. 12. Wang, P., Chao, K.M., & Lo, C.C. (2010). On Acknowledgements optimal decision for QoS-aware composite service selection. Expert Systems with Applications , Vol. This work was partially supported by the Spanish 37, No. 1, pp. 440–449. project TIN2013-44641-P. Oscar Cabrera Bejar is 13. Mohanty, R., Ravi, V., & Patra, M.R. (2010). Web- a Ph.D. student at the UPC using a CONACYT services classification using intelligent techniques. grant. Expert Systems with Applications , Vol. 37, No. 7, pp. 5484–5490. References 14. Tao, Q., Chang, H.Y., Gu, C.Q., & Yi, Y. (2012). A novel prediction approach for trustworthy QoS of 1. Taylor, S., Iqbal, M., & Nieves, M. (2011). ITIL web services. Expert Systems with Applications , Version 3 Service Strategy. The Office of Vol. 39, No. 3, pp. 3676–3681. Government Commerce. 15. Cai, H., Hu, X., Lü, Q., & Cao, Q. (2009). A novel 2. Papazoglou, M. (2007). Web Services: Principles intelligent service selection algorithm and and Technology. Pearson-Prentice Hall. application for ubiquitous web services 3. Menasce, D. (2002 ). QoS issues in web services. environment. Expert Systems with Applications , Internet Computing, IEEE, Vol. 6, No. 6, pp. 72–75. Vol. 36, No. 2, pp. 2200–2212. 4. Robertson, S. & Robertson, J. (2012 ). Mastering 16. Sha, L., Shaozhong, G., Xin, C., & Mingjing, L. the Requirements Process: Getting Requirements (2009). A QoS based web service selection model. Right. Pearson Education. International Forum on Information Technology
Computación y Sistemas, Vol. 18, No. 4, 2014, pp. 665–682 ISSN 1405-5546 doi: 10.13053/CyS-18-4-2057 682 Oscar Cabrera, Marc Oriol, Xavier Franch, Jordi Marco, Lidia López, Olivia Graciela Fragoso Díaz, and René Santaolaya
and Applications 2009, IFITA'09, IEEE, pp. 353– Marc Oriol is a Ph.D. student in Computer 356. Science at the Universitat Politècnica de 17. Alrifai, M., Risse, T., Dolog, P., & Nejdl, W. Catalunya, UPC, Barcelona, Spain. He obtained (2009). A scalable approach for QoS-based web his M.Sc. degree in Computing from the same service selection. Service-Oriented Computing university. He is a member of the GESSI (ICSOC’08) Workshops, Springer, pp. 190–199. research group at UPC. His current research 18. Huang, A.F., Lan, C.W., & Yang, S.J. (2009). An lines include service-oriented computing, quality- optimal QoS-based Web service selection scheme. of-service, and monitoring. Information Sciences , Vol. 179, No. 19, pp. 3309– 3322. Xavier Franch is Associate Professor and Head 19. Lin, C.F., Sheu, R.K., Chang, Y.S., & Yuan, S.M. of the GESSI research group at the Universitat (2011). A relaxable service selection algorithm for Politècnica de Catalunya, UPC, Barcelona, QoS-based web service composition. Information Spain. He obtained his Ph.D. and M.Sc. in and Software Technology , Vol. 53, No. 12, pp. Informatics from this University. His current 1370–1381. research lines include service-oriented 20. Gao, Z.P., Chen, J., Qiu, X.S., & Meng, L.M. computing, requirements engineering, software (2009). QoE/QoS driven simulated annealing- quality, and software architecture, among others. based genetic algorithm for Web services selection. The Journal of China Universities of Jordi Marco is Associate Professor and member Posts and Telecommunications , Vol. 16, pp. 102– of the GESSI research group at the Universitat 107. Politècnica de Catalunya, UPC, Barcelona, 21. Salton, G., Wong, A., & Yang, C.S. (1975). A Spain. He obtained his Ph.D. and M.Sc. in vector space model for automatic indexing. Computing from this University. His current Communications of the ACM, Vol. 18, No. 11, pp. research lines include service-oriented 613–620. computing, conceptual modeling, container 22. Barba Romero, S. & Pomerol, J. (2000). libraries, and computer graphics. Multicriterion Decision in Management. Principles and Practice . Kluwer Academic Publishers. Lidia López is a researcher of the GESSI 23. Peña, V.H., Lai, T.L., & Shao, Q.M. (2008). Self- research group at the Universitat Politècnica de normalized processes: Limit theory and Statistical Catalunya, UPC, Barcelona, Spain. She obtained Applications . Springer. her Ph.D. and Engineer degree in Informatics 24. Knappe, R. (2005). Measures of semantic from this University. Her current research lines similarity and relatedness for use in ontology- include service-oriented computing, goal-oriented based information retrieval . Doctoral dissertation, modeling, and open source software. Roskilde University. Olivia Graciela Fragoso Díaz is a researcher in 25. Bernstein, A., Kaufmann, E., Kiefer, C., & Bürki, the Software Engineering area at the National C. (2005). Simpack: A generic java library for similarity measures in ontologies . University of Center for Research and Technological Zurich. Development (CENIDET), Cuernavaca, Morelos, México. Her areas of interest are web services Oscar Cabrera is a Ph.D. student in Computer selection, web services for e-learning, software Science at the Universitat Politècnica de reusability, and processes for software Catalunya, UPC, Barcelona, Spain. He obtained development. his M.Sc. degree in Computing in the Software René Santaolaya Salgado is a researcher in the Engineering area from the National Center for Software Engineering area at the National Center Research and Technological Development for Research and technological development (CENIDET), Cuernavaca, Morelos, Mexico. He is (CENIDET), Cuernavaca, Morelos, México. His a member of the GESSI research group at UPC. areas of interest are web services, software His current research lines include service- reusability, and integrated environments for oriented computing, current trends in smart cities, software development visual programming. context modeling, quality models, and information technology in software development. Article received on 10/08/2014; accepted on 01/11/2014.
Computación y Sistemas, Vol. 18, No. 4, 2014, pp. 665–682 ISSN 1405-5546 doi: 10.13053/CyS-18-4-2057