Description, Classification and Discovery Approachesfor Software Components: A Comparative Study

Introduction

Component-based and service-oriented software architectures are likely to become widely used technologies in the future distributed system development. Component reuse is a crucial requirement for the development process of component-based and service-oriented software.

Components are developed as important and big autonomous and customizable software units. The successfulness of the reuse is important and depends on the efficiency of the search procedure. The search step is essential in the development process, since the developer is generally faced with a significant number of various component types. The search step may fail if the explored component repositories are not appropriately structured or if the required and the provided services are not correctly compared. The use of a component repository, having a well-defined structure, is crucial for the efficiency of the CBD approach. This allows the developer to easily seek and select the component which perfectly meets his/her needs.

Through this study, we analyze the key factors that are necessary for obtaining a well-organized software component repository and software components having a pertinent description for the search procedures (see Figure 1). These factors act not only on the precision of the specified request but also on the component resulting from the search process.

Figure 1.

For component description, two generation approaches of description are distinguished:

Manual generation (Erdur & Dikenelli, 2002) and automatic generation. The automatic generation is relying on different methods such as introspection (Sessions, 1998), (ONeil & Schildt, 1998), trace assertion (Whaley et al., 2002) and invariant detection (Perkins & Ernst, 2004).

The second part identifies and describes five categories of methods for representing component classification. The first is the adhoc method, called also behavioral method (Podgurski & Pierce, 1992), (Atkinson & Duke, 1995). The second is based on the semantic characteristics of the component (Penix & Alexander, 1999). The third uses the facet classification (Damiani et al., 1999), (Ostertag et al., 1992), (Vitharana et al., 2003), (Ferreira & Lucena, 2001). The fourth method is based on the lattice (Fischer, 2000). Finally, the fifth method applies the notion of ontology (Erdur & Dikenelli, 2002), (Meling et al., 2000) to describe and classify components. Different techniques are used to organize components in repository: the cluster technique (Nakkrasae et al., 2004), the thesaurus technique (Liao et al., 1997) and the subsumption technique (Napoli, 1992).

The third part addresses the component discovery techniques related to classification methods. A successful adequation between the description and the classification methods should provide a powerful discovery service. This allows the developer to easily discover the appropriate component that meets his/her needs. The most popular discovery techniques are based on: genetic algorithms (Xie et al., 2004), neural networks (Nakkrasae et al., 2004), symbolic learning (Utgoff, 1989) and probabilistic information retrieval (Yunwen & Fischer, 2001). These techniques use the decision tree algorithm (Ruggieri, 2004), (Vasiliu et al., 2004) or unification of the component description in the comparison phase (Yao & Etzkorn, 2004).