Discovering configuration workflows from existing logs using process mining

Abstract

Variability models are used to build configurators, for guiding users through the configuration process to reach the desired setting that fulfils user requirements. The same variability model can be used to design different configurators employing different techniques. One of the design options that can change in a configurator is the configuration workflow, i.e., the order and sequence in which the different configuration elements are presented to the configuration stakeholders. When developing a configurator, a challenge is to decide the configuration workflow that better suits stakeholders according to previous configurations. For example, when configuring a Linux distribution the configuration process starts by choosing the network or the graphic card and then, other packages concerning a given sequence. In this paper, we present COnfiguration workfLOw proceSS mIning (COLOSSI), a framework that can automatically assist determining the configuration workflow that better fits the configuration logs generated by user activities given a set of logs of previous configurations and a variability model. COLOSSI is based on process discovery, commonly used in the process mining area, with an adaptation to configuration contexts. Derived from the possible complexity of both logs and the discovered processes, often, it is necessary to divide the traces into small ones. This provides an easier configuration workflow to be understood and followed by the user during the configuration process. In this paper, we apply and compare four different techniques for the traces clustering: greedy, backtracking, genetic and hierarchical algorithms. Our proposal is validated in three different scenarios, to show its feasibility, an ERP configuration, a Smart Farming, and a Computer Configuration. Furthermore, we open the door to new applications of process mining techniques in different areas of software product line engineering along with the necessity to apply clustering techniques for the trace preparation in the context of configuration workflows.

Evolvable SPL management with partial knowledge

“…The SPL progressively supports capitalizing on what is not explicit by reasoning about these configurations. Then the main issue is not to determine the configuration workflow that best suits the actors according to the previous configurations [45], but to guide them in composing a solution for an unprecedentedly studied problem. Concurrently, it is not a question of generating random samples [23], whose relevance could not be precisely verified (e.g., stuffing the SPL with all the available algorithms and pre-processing components from the literature).…”