Use of clustering methods to solve the problem of identifying configuration items in IT project

The object of the research is the IT project configuration management process.

During the research, the task of identifying the configuration items (CI) of the IT product has been solved. Research in this field is mainly aimed at solving the problem of configuration analysis during the refactoring of a monolithic IT product into separate services or microservices. The question of decomposition methods of the description of the architecture of the developed IT product into separate functional SIs remains practically unexplored.

In the course of the research, it has been proposed to use hierarchical and non-hierarchical clustering methods to solve the problem of identifying functional CIs. As examples of hierarchical clustering methods, the agglomerative clustering method (AGNES using the nearest neighbor algorithm) and the divisive clustering method DIANA have been proposed. The k-means algorithm has been proposed as an example of non-hierarchical clustering methods. In addition to them, it has been suggested to use one of the grapho-analytic clustering methods for comparison, which was developed to solve the decomposition problem of the description of the monolithic architecture of the software product into separate microservices.

The starting data for the research is the description of the architecture of the "Formation and management of the individual plan of the scientific and pedagogical worker of the department" functional task at the level of individual functions. 10 functions of this problem have been considered as CI. Descriptions of 12 entities of the problem database have been used to define these functions. The features of the solution have been considered and the results of the solution to the problem of identifying functional CIs using all four selected clustering methods have been obtained.

A comparative analysis of the process of solving and the obtained results of solving the task of identifying functional CIs using all four selected clustering methods has been carried out. It has been established that the best alternative is to use hierarchical clustering methods to solve this problem. This makes it possible to further consider the task of assigning to individual teams of IT project executors a list of functional CIs that require implementation as a sequence of individual single-criteria optimization tasks.