• Aucun résultat trouvé

Existing Solutions for Web Service Selection

Dans le document Advanced Information and Knowledge Processing (Page 176-181)

A number of studies for web service selection have been carried out. One of the most well known techniques is “matchmaking”. It is employed in the situation where ser-vices with semantic descriptions for their functional attributes are available from an

Internet search. Several service matchmaking techniques [7,15,42] have been devel-oped to meet the needs of both consumers and providers as illustrated in Figure 7.2.

Fig. 7.2 Requirements matchmaking between service customer and service provider

Ran [33] proposed a new QoS-based service registration and discovery model via exchanging SOAP messages [37] to explore the possibility of QoS being involved in UDDI registry information [39]. In this model, service providers have to send QoS claims to QoS certifiers, responding to third party or forum web services, for certi-fication. The service customer is responsible for verifying QoS claims. The QoS in-formation finally will be registered in the UDDI registry associated with the function description, once QoS claims have passed QoS certifier verification. However, QoS of certified services generally are dynamic in nature on the Internet. This might be taken into account through continuous monitoring and checking. Hence we involve adding two checking items, QoS monitoring and opinions feedback from customers into Ran’s model, and the original model is modified as shown in Figure 7.3.

The new UDDI registration mechanisms help customers to discover and locate the required service by looking up a WSDL document as well as certified QoS [32].

Moreover, consensus of service consumers on QoS attributes has to be considered for the web service QoS certifier in the QoS computation process. Balke and Wag-ner [3] introduced the “cooperative discovery” concept for evaluating web services in detail which comprises three phases of interaction with services, namely, (i) ser-vice discovery, (ii) serser-vice selection, and (iii) serser-vice execution. Based on Figure 7.1, we reorganized the three phases as shown in Table 7.2, which specifies the extensive definition for selection of QoS-aware web services provisioning.

QoS
 computation


Serach
and
 find
and



QoS
monitoring
 Vertify
QoS


QoS
ranking
 Request
and
 respond


SOAP
and
XML


WSDL,certified
QoS
 Register

 WSDL,certified
QoS


Opinion
feedback


Certifty
QoS
 Web
service


QoS
certifier

New UDDI registry Web service

consumer

Web service provider

Fig. 7.3 QoS-centric web registration and discovery model Table 7.2 QoS-aware web services discovery and selection

Phases Tasks Task description Support tools Dealer

Service Function definition Specify the terms of WS WSDL, OWL-S WS provider registry functionalities using ontology

language or WSDL

Service registry Register and receive a official UDDI database WS provider ID for applied service to

publish to the Internet

Service QoS certification Accept and certify the QoS assessment QoS certifier

discovery application of service model

QoS attributes

Service Announce the features of WS Business portal WS provider advertisement

Service discovery Perform and find the related Browser WS consumer services based on a user’s

request

Service selection Select one of the desired service Browser WS consumer Service Service execution Carry out service binding Browser and WS consumer,

execution and execution service WS provider

QoS Collect customer opinions to Browser and WS consumer monitoring QoS certifier for reflecting QoS opinion

user expectation database

For emerging B2B businesses, selected services are aggregated to form compos-ite services. A composcompos-ite service is a service produced by a composition of other services to complete the desired service activities. For example, a consumer may wish to discover a composite service containing flight booking, restaurant reserva-tion, and renting a car. Zeng et al. [42, 43] addressed the issue of selecting web services by maximizing user satisfaction expressed as utility functions over QoS attributes; this selection model considered multiple criteria such as price, duration, reliability in which budget constraints and preferences set by the user. This approach is a global planning method so it can optimally select component services by lin-ear programming techniques. For the example of a composite service as a travel planner, it aggregates multiple component services for flight booking, travel insur-ance, accommodation booking, and car rental, which can be executed sequentially or concurrently, as illustrated in Figures 7.4 and 7.5.

Customer
 Requirements


Hotel
Services
(S8~
S15)
 Insurance
 Services
 (S3~

S7)


Flight
Services(S1~
S2)


Car
Rental
Services
(S16~
S17)
 Fig. 7.4 The discovery and selection of composite services

S4 S11 S16

S1

S5 S7

S15 S17

start
 end


Fig. 7.5 A possible plan of selected composite services

Sirin et al. [36] developed a goal-oriented and interactive composition approach that uses matchmaking algorithms to help users filter and select services while build-ing their composition service. The matches are filtered usbuild-ing ontological reasonbuild-ing on the semantic descriptions of the services. They developed a prototype on the ba-sis of these ideas to test the system by generating OWL-S descriptions for some

of the common web services. Each composition the user generates via the existing prototype will be realized as an OWL-S Composite Process, meaning that it can also be advertised, discovered, and composed with other services. They adopted a quality model considering five generic quality criteria for elementary services:

(1) execution price, (2) execution duration, (3) reputation, (4) reliability, and (5) availability. The global service selection is executed with a set of execution plans, P= (p1,p2, . . . ,pn), wherenis the number of plans. After a set of execution plans is generated, the scheme selects an optimal execution plan using Simple Additive Weighting (SAW). This work pointed out that the accuracy of the matches found by the inference engine depends on how detailed the ontologies are. Richer ontologies with more specific descriptions for sensors and their nonfunctional properties will help the engine find better answers to the queries.

Zhou et al. [45] discriminated between functional and non-functional QoS prop-erties of web services, where functional propprop-erties can be measured in terms of throughput, latency, response time; non-functional properties address of various is-sues including integrity, reliability, availability and security of web services. Well-defined metrics are utilized by measurement organizations to monitor and evalu-ate the promised service level objectives. A match-making prototype is designed to prove the feasibility of the approach. For the match between request and advertise-ment, there are five types of match possibilities: subsume, exact, plugin, intersection, and disjoint ranging from the best matching degree to the worst matching degree, respectively. When the service provider publishes their service QoS profile through the publish interface, the ontology will be parsed. If the parsing process ends suc-cessfully, the ontology is stored in the server’s repository and then is rendered into a description kept in the knowledge base. By classifying on its knowledge base, the Racer engine organizes the ontologies’ taxonomy. When the service requester sub-mits an inquiry, the matchmaker will return the subsume, exact, plugin, and intersec-tion matching list respectively. The prototype demonstrated that the matchmaking is suitable for small or middle sized advertisement repositories.

Lin et al. [25] and Liu et al. [27] treated the selection of QoS-driven web services with dynamic composition as a fuzzy constraint satisfaction problem and applied an optimal search approach with adjustments to service composition. They consider three generic quality criteria which can be measured objectively for elementary ser-vices: (i) execution price, (ii) execution duration, and (iii) reputation. Compared to Sirin et al.’s approach [36], criteria such as availability and reliability are not included in the model, due to the use of active user feedback and execution mon-itoring. The reputation of a service is a measure of its trustworthiness. It mainly depends on end user’s experiences of using the service. To demonstrate the pro-posed QoS model, they implemented a QoS registry within a hypothetical phone service (UPS) provisioning market place that is implemented using BEA Weblogic Workshop toolkit. It consists of various service providers who can register to pro-vide various types of phone services such as long distance and local phone services, wireless and broadband. The UPS marketplace has web interfaces which allow a customer to login and search for phone services based on his/her preferences. For

example, the customer can specify whether the search for a particular type of service should be price or service sensitive.

In summary, service selection approaches [15, 27, 36, 43, 45] can be mainly di-vided into two categories: Multiple Attribute Decision Making (MADM) [8,41] and mathematical programming. MADM methods [15, 27, 45] concentrate on that QoS attributes can be collected and enforced objectively, then the traditional MADM the-ory can be applied to obtain a consistent ranking of service alternatives. Mathemati-cal programming methods [36, 43] comprise linear programming (a single objective function) and multiple goal programming. It concerns about interactive composition selection that uses preset planning to optimally select component services during the execution of a composite service.

The “matchmaking” approach, however, relies on the advertisements from ser-vice providers’ subjective views that could lead to divergent perception between consumers and providers. Consumer expectations and their common preferences (i.e., consensus) on QoS should be considered in the process of service selection.

To ensure the consensus between consumers and providers, Lin et al. [26] proposed a QoS Consensus Moderation Approach (QCMA) in order to perform QoS ratings based on [13, 21] in order to alleviate the differences in QoS characteristics.

The aforementioned methods advanced knowledge in QoS-aware service discov-ery and selection, but nevertheless, there remains the following significant issues for debate: (i) the perception of QoS attributes needs to adjust according to consumers’

preferences, (ii) how to determine weights (importance) of QoS attributes, and (iii) the ranking order of service alternatives should be decided on the basis of group consensus. To enable effective QoS-aware service selection, a new web service se-lection model is proposed, which includes the following important aspects.

• Imprecise preference: this model should be able to handle vague preferences or linguistic opinions for QoS attributes expressed by service consumers in the pro-cess of selecting web services.

• Be able to explore the optimal solution weighting of QoS attributes.

• Consensus-centric service ranking: the approach should be capable of realisti-cally gaining a consensual ranking on web service alternatives according to con-sistence and inconcon-sistence measurements of performance ratings.

• Inspired by Li and Yang’s work [24], we extend our previous work [40], to select QoS-aware web services using fuzzy linear programming techniques by min-imizing the inconsistency measurement. More detailed information about this model is described in the next section.

Dans le document Advanced Information and Knowledge Processing (Page 176-181)