Background, previous co-operation related to the project

One of the key questions in software development is software size estimation. Software size is a primitive measure from which other important values are calculated (e.g. effort, productivity and duration). These values are particularly interesting for project managers at the beginning of a project. In practice, an intuitive approach is often used, not a methodological one. The results are often unsatisfactory. The efficiency of the intuitive estimates can be indirectly measured via software project success factors, especially time and budget overruns.

For systematic software size estimation, different methods are used, all of which have their roots in the Function Point Analysis (FPA) method. Albrecht introduced this method in 1979. Since then it became the most important method for software size estimation. The method introduced a specific way of representing a software system and distinguished between data functions and transactional functions. The method was intended for all domains, although in practice, its accuracy is different within different domains. A more detailed empirical analysis of the method revealed some weaknesses, which include:

Correlation between elements of the FPA method.

The unsuitability of the Value Adjusment Factor (VAF) and General System Characteristic (GSC).

The violation of the monotony axiom.

From a practical standpoint, it can be concluded that the FPA method application is more difficult with object-oriented projects. The elements and constructs of the FPA method are not directly applicable to object-oriented concepts. Therefore, a mapping of object-oriented concepts into the FPA elements is needed. The mapping is not defined within the FPA method itself and is consequently not uniform. Different authors have proposed different mapping functions, mostly in the form of additional rules. Information is gathered from different diagrams (e.g. Use Case diagrams, class diagrams, sequence diagrams) which are considered separately. More detailed research has shown that the weight factors of the standard FPA method have to be calibrated for use in object-oriented projects. Consequently, several FPA-like methods were developed that map object concepts into metrics similar to function points. Sneed proposed Object Points and Minkiewicz developed Predictive Object Points. The results are therefore incomparable to those calculated within the original FPA method. Additional adjustment factors are needed that have to be proven statistically.

Project managers need to estimate the effort needed for a project’s completion and a project’s duration. Most methods are applicable during the late analysis phase or the beginning of design, which has little true value for project management activities.

Most of the current methods for software size estimation have the same roots (the FPA method), but lack the formal model needed for method comparison. In the past, a few attempts were made to establish such a formal model, however none have become predominant in practice.

We had preparatory discussions with the Slovenian partner at Microsoft Academic Days for Central and Eastern Europe, November 16th-19th in Prague.




Reasons of co-operation (with special regard to the complimentary character of the research groups)

Main objectives are related to the definition and development of strategies, approaches and tools to support:

Identification of minimum set of metrics needed to establish valuable metrics program for a given project.

The UML models definition, required in each development phase to be able to estimate software size in desired accuracy.

Definition of correlation factors between function points and source-code metrics for .NET environment.

The framework for tool implementation that support data collection and size estimation for .NET environment using UML models in Visio.

Existing complementary work of both research groups presents solid ground for joint efforts in function points and metrics for .NET environment - see references.

Objectives of the project and planned utilisation of the results

The focus of this project would be a practical application of metrics suitable for object, component and aspect -oriented development in specifics target environment namely Microsoft .NET. The concrete environment is needed to build a bridge between two metrics groups - source code metrics on one side and high level project metrics on the other side. Within the project a subgroup of source code metrics that has a greatest impact and higher correlation coefficient with high level metrics for estimating project size will be defined collected and tested to formally prove their importance. In a more general sense the project will define a framework for software metrics plan that can be used within software companies and further applied on .NET environment. The framework itself will focus on specific metrics used within software projects to estimate project size. However the results and the framework itself are expected to be applicable to other metrics families as well. For the practical part of the project conceptual definition of architecture will be established that will enable the end user to easily compose a tool for metrics gathering and analysis in .NET environment. The analysis will be founded on UML models composed in Visio. The UML models will then be used to calculate project size in function points. The correctness of results will be statistically proven using actual hours; further on the correlation between function points and selected source-code metrics for .NET environment will be investigated.

Expected advantages of the co-operation for Hungary, as well as for the partner country

Common result is the possibility of development practical tools which are easy-to-use, controllable and customizable and applicable in software development based on function-point and source-code metrics. Beside the Hungarian partner widen its knowledge on function-point mappings and project estimations. Similar result is expected for the Slovenian partner in the field of source-code metrics.

Assignments for both the Hungarian and the foreign institutions

Based on the previous research directions investigations in function-point metrics and development of theoretical framework of the whole project will be controlled by the Slovenian partner,. Development, testing and evaluation of source-code metrics are planned as the role of the Hungarian partner.

Time schedule (monitoring points)
The project will have two main phases:


Phase I.

1. Comparison of existing mappings from UML diagrams into FPA method elements for .NET environment
Existing mappings from different authors will be compared, common principles and deviations identified.

2. Definition of the meta-model for describing FSM methods and selection of the most suitable method for .NET environment

Existing FSM methods are different in many ways. The element names, calculation function and procedure are different. Therefore, a comparison of different methods is difficult. The purpose of this step is defining the meta-model and the corresponding mathematical model. With a comparison of procedures, greater accuracy of estimates and quality of mapping functions will be possible. According to the results the most suitable FSM method for the target environment will be selected.

3. Integration and update mapping of object concepts into FPA method elements using UML 2.0 standard
A detailed comparison of mappings is expected to show their advantages and disadvantages in practical use. With the integration of different procedures and the definition of additional rules, we will develop an original approach for the software size estimation of object-based systems.

Phase 2.

4. Development of components for counting automation
Since mapping defined during this research will be based on formal models, it is expected that the automation of counting steps for .NET environment will be possible. The component will map software systems expressed with UML diagrams composed in the Visio tool into size in function points. The counting procedure will be partial or fully automated according to the level of system abstraction. It is expected that automation will lower the subjectivity error and consequently improve accuracy.

5. Integration with the source code metrics
It is important that the method defines the source-code metrics data necessary for results comparison and achievement of desired accuracy level in later projects. The latter depends on the quality of the input data and selected subset of metrics. Therefore, the source-code metrics collection should be fully automated and correlation between function points and each selected method clarified.

6. Solution evaluation
The proposed solution will be evaluated and statistically tested on actual .NET projects.