An ontological approach for managing and selecting QoS Web services

3.1 Quality of service ontologies applied to Web services

Tran and Tsuhi [7] have made a review of the current approaches on the topic of quality of services (QoS) focused on the development of ontologies applied to Web services. Different types of QoS properties can have different types of values, moreover they can have different trends: they can be positives (the higher the value the higher the benefit) or negatives (the higher the value the lower the benefit). The referred review lists the following approaches:

OWL-Q is an ontology of superior level whose modular focus make it flexible to add support ontologies. This is an upper level ontology that extends OWL-S, and it is separated in multiple and independent facets. Connecting Facet links OWL-Q and OWLS-S ontologies, defines concepts and vocabulary for QoS advertisements and requirements. Basic Facet associates a service profile with QoS offers. QoS Metric Facet describes classes and properties used for defining QoS metrics. Unit Facet describes the unit of a QoS metric. This ontology provides a detailed specification for modeling QoS information. Its modular approach makes this ontology flexible to add more necessary support ontologies for an applied application.

The ontology proposed by Maximilien and Singh [14] enables to agents of its proposed service algorithm finding coincidences with published quality levels for consumers with specific preferences on quality of services. Providers state politics and consumers state preferences using the ontology. The ontology is divided in three parts. The upper ontology which contains generic concepts of quality and defines the basic concepts associated to quality like measure and relations. The middle ontology incorporates many aspects of quality found in distributed systems like availability. Tipically, the middle ontology is completed with a lower ontology specific to the domain.

3.2 Selection of Web services

Maximilien and Singh [14] implemented an architecture based on agents which is executed in a WSAF (Web Service Agent Framework). WSAF incorporates agents of services selection which use a QoS ontology and an XML language of politics allowing to consumers and providers of services to expose their quality parameters. For each service, WSAF creates a service agent which exhibits the service interface with additional functionality to capture the preferences or politics of services quality and consulting other agents about coincidences. The proposed QoS ontology enables to service agents to find coincidences with quality levels published for consumers with specific preferences of quality of services. The referenced work does not give enough detail about the executed tests.

Junhao et al. [8] propose an ontology which supports different measurements and the matching of heterogeneous quality of service attributes. It also has a great capability of extensibility and semantic description. Based on this ontology, an algorithm of semantic Web service selection was developed consisting in four phases: first, a semantic matching is done between an array of service requests and a set of published services, then the measures of quality of service parameters are unified. After, a pairing of values with the set of published services is done permitting, as last phase, a user to select the best Web service according to a custom selection. The results of the realized experimentation showed that the proposed algorithm selects the service which is the best adapted to the requirements.

Raj and Sasipraba [15] present a Web service selection model in which a quality of service manager is implemented. The manager acts as mediator between the client and the provider. It binds the published data with the service. The selection is performed based on the functional requirements and the quality of service of the client which specifies threshold values of quality parameters. After that process, a set of service candidates satisfying functional and quality requirements is obtained for being classified and selected. Classification is carried out according to quality of service usefulness. Finally, the service with the best quality is selected and returned to the client. The experimentation results show that the Web services were succesfully published and classified according to user requirements.

Tran, Puntheranurak y Tsuji [6] acomplish a coincidence of services by means of the evaluation on how a request and a service fulfill the requirements for both of them. The role of the algorithm is assessing the level of functional satisfaction of a Web service according to a request by pairing their functional descriptions for the service. If it exists more than one operation of service that have coincidence with the requested operation the list of operations is ordered in agreement with a set of criteria which take into account the degree of fulfillment with the requirements of both the request and the services. Although it is mentioned that the algorithm was implemented no results are mentioned.

Chitra et al. [10] proposed a model for selection of Web services using Naïve Bayes to classify the quality of service by means of the pairing of ontologies’ patterns. Naïve Bayes is an algorithm of machine learning which works on the conditional probability of Bayes theorem. The user requests a service with or without specifying QoS functionalities. A Web service that coincides with the required features is obtained from a UDDI register. All the obtained services are converted to ontologies and stored for future reference. The pairing process is performed in three steps: syntactic similarity, based on string comparison, semantic, computed for words using WordNet and structural which compares the structure of ontologies through a number representing a similarity level between concepts. Tests done in a scenario involving real Web services show that a better selection can be achieved with the proposed method.

Lim et al. [16] propose an approach of selection for Web services based in the satisfaction of three parts: user, Web service and community, taking into account that the Web services reside in a community. For the first part, the parameters of quality of service for the current service. In the second part, because of being part of a community the service expects certain workload, a satisfied Web service will rest available in the community for longer and will distribute its resources to a greater extent. An unsatisfied service could leave the community earlier and negatively affect the level of quality of service for the users. Finally, the selection is carried out through a master Web service which measures its satisfaction in terms of cost, such satisfaction represents how good the community is respect to other communities. The realized experimentation shows that the consideration of the three parts has beneficial results for the user compared with the approach that considers only user satisfaction.

Shi et al. [17] present a user centered calculation of quality of service values, taking into account differences between experienced and novel users. The experienced users can have an idea of the range for a particular quality of service parameter and also to have a preference on some parameters rather than others. To apply such knowledge is impossible for new users. The calculation of the quality of service parameters is separated in three phases: preprocessing of values (like normalization or other process), user satisfaction (degree of coincidence between the required quality by the user and the quality provided by the available Web services) and the agreggation of multiple properties (different weights for each property). For the study in question a translation service is selected and according to the cost and the quality of the translation the obtained results were satisfactory for both new and experimented users.

Nagy et al. [18] proposed an algorithm of selection based on quality of service which is able to identify the best Web service candidate given the service description and the criteria for quality of service provided by the user. A semantic matcher is used to improve the selection precision. Data storage techniques are used to realize a better selection based on requirements and previous selections done by the user. The algorithm obtains a set with the best services already classified. Using as criteria of quality of service the level of confidence, execution time and the cost by service in a set of simulated services. The tests showed that the proposed method provides a bigger number of candidate services extending the options to find the best service, though the bigger the number of providers the bigger the planning time for requests.

Harshavardhanan et al. [19] proposed an architecture of quality of service managers to select the best Web service which is obtained from a query realized by the client including his requirements of quality of service. The proposal employs a UDDI register to accomodate the information of quality of service and the quality agent to ease the service discovery. The quality agent will select the most suitable of the Web services for the requester by satisfying his constraints and preferences of quality of service. The agent architecture is divided in three layers: the first is the interaction one which contains the clients and services providers, the second is the agent one which contains the methods of selection of Web services and also the methods of service validation (generally carried out by a third party), and finally, the repository of Web services where the services and the quality of service parameters are published.

3.3 Limitations of related work

The ontologies reviewed in Section 3.1 have the following limitations [7]:

OWL-Q has a weak support of priorities for quality of service and does not include common definitions of QoS.

The ontology proposed by Maximilien y Singh [14] does not support metrics nor priorities although it covers the most of the other limitations. The ontology provides a great variety of classes covering many aspects of QoS parameters exceeding the needs of the work presented in this paper. We have developed a new ontology taking as starting point the referenced ontology, details are presented in Section 4.2.

In the aspect of service selection, the reviewed works, in general, do not consider that the user requirements can change during the process [8], they only return a service as a result and therefore they are not prepared for a composition context [9]. In [6, 14] implementations are introduced but without details or results.

In order to register the Web services and their respectives quality of service parameters, two options were considered:

A modified UDDI register for the support of the semantic information. Universal Description, Discovery and Integration (UDDI) [20] is a set of registers based on Web which expose information of companies or other entities together with their technical interfaces or APIs. The registers are available through operator sites, and anyone can seek information about Web services published by or in behalf the company. However, the lack of understandable by machines semantic information in the UDDI registers hinders its use for automatic services discovery. In order to overcome these limitations tools like FUSION [21] have been developed. FUSION is a register of Web services semantically improved based on UDDI specifications which improves the publishing of Web services.

4.2 Ontology for Web services

An ontology will be used in two moments of the composition process. First, when obtaining and storing QoS parameters values, and then to the moment of recovering them and to determine which services will be used in further stages of the process. Implemented in Java, the conection with the ontology is made by using the platform Jena. This platform provides functions to access, query and update by means of the query language SPARQL the information stored in the ontology. Such information is found in a set of independent files which can be modified without changing the original file for the ontology. Figure 1 shows the needed SPARQL code to add a new value for a parameter of QoS in a specific Web service. Additionally it can be added data like date among others. Figure 2 shows the code to retrieve the value of the availability parameter for a specific Web service. Figure 3 shows an image in Protégé of the classes that define our ontology. The Tables 1 and 2 show the elements belonging to the ontology and their description. The Jena’s API is available at https://jena.apache.org/documentation/javadoc/.

OPEN IN VIEWER

Fig.1

Code in SPARQL to insert a new value for the parameter of availability in the service represented by ServiceId.

OPEN IN VIEWER

Fig.2

Code in SPARQL to retrieve the values for the parameter of availability in the service represented by ServiceId.

OPEN IN VIEWER

Fig.3

Developed ontology for Web services registration.

6.1 Evaluation agent of quality of service

In order to obtain the values of QoS parameters, two classes of WS were considered: The first class is a weather service in which the temperature of a city can be obtained. The another class is a Web service that can be used to convert temperatures from one unit to another, for example from Fahrenheit to Celsius. In the class of weather we have 3 Web services. These were selected because they were freely available and had a clear yet enough documentation that allowed the use of services without major complications. Of course, the strategy defined here can be extended, applying it to others Web services with diverse characteristics, such as greater complexity, greater number of suppliers, paid ones as Amazon Web services, etc.

National Oceanic and Atmospheric Administration Web service.1

In the class of temperature conversion there are 4 Web services.

A simple calculator Web service like the one available in http://www.dneonline.com/calculator.asm. Given that C = 5 * ( F - 32 ) 9 ,(5) We can use subtraction, multiplication and division operations for temperature conversion.

While weather Web services can provide the temperature in the desired unit, for this work it is assumed that they will return the result in Fahrenheit degrees. Likewise, temperature conversion Web services (except the calculator) can be indicated on which unit the response is to be found. It is assumed that they will receive the temperature in Fahrenheit degrees and return it in Celsius degrees.

The process to get the value of a quality of service parameter was as follows: Each WS was executed a total of C _T  = 50 times, recording the total time of all requests (the time of each request from the instant when the Web service is called until the response is received), and the number of successful and failed requests. Using the formulas presented in Section 4 for the Availability, Response Time and Throughput parameters, the values of each parameter were obtained for each WS. The process was repeated at instants when a real request could be made. Some requests were made in the morning and others in the afternoon, for several days (approximately one week). A pause was done for a while (few days) and then the process was repeated.

In general, WS were available on all times when requests were made. Only the ConvertTemperature service had availability problems, and if it has an availability of 0 also has a throughput of 0.

Table 3 shows average values of QoS parameters obtained for each WS provider.

Experimentation with the Web services selection agent consisted in applying equations such as 4 with different weights to determine if a Web service will be taken into account by the composition manager. The Equation 6[27] were applied to the obtained QoS values: QoS ( s ) = ∑ W m × q i + ∑ W n × 1 q j(6)

Where W _m and W _n are the weights of the quality of service parameters, q _i represents the negative QoS parameters (better if it has a lower value) and q _j represents the positive parameters.

The composition manager consists of joining the Web services according to the values obtained by the equations using the Dijkstra algorithm.7⁷ Therefore, the service with the lowest value QoS (s) would be selected by the composition manager. However, a modification was made aiming at the service with highest value QoS (s) is selected, in such case q _i represents to the positive parameters and q _j to the negative ones.

The experiments are aimed at verifying whether the application of equations affects the composition process.

In the real scenario, both equations were used to calculate the number or reward representing the Web services and the simple Dijkstra algorithm was applied in the composition manager.

Although the use of any of two equations did not affect the final result, each one has the following disadvantages:

In Equation 4, the use of a very large weight for the response time parameter causes the number obtained to be negative, which is inadequate for the composition manager, since the latter can not handle negative weights. This was the case of the set of weights marked with the number 2 in Table 4.

For the artificial scenario, the same method of the real scenario was used, except that only sets of weights numbers 1 and 3 were used. In this case, the results indicated that the use of one or the other equation affected the final result of the composition manager. 100 layers of 15 WS each were used. The results obtained are as follows.

Taking into account the result of the composition manager obtained with the Equation 6, and the rewards obtained with the Equation 4, and comparing it with the result obtained with the latter, it is concluded that, it is precisely with the result obtained with Equation 4 that a better quality of service value is obtained. In addition, a review layer by layer showed that the first equation selects better Web services.