Graph-based implicit knowledge discovery from architecture change logs

Abstract

Service architectures continuously evolve as a consequence of frequent business and technical change cycles. Architecture change log data represents a source of evolution-centric information in terms of intent, scope and operationalisation to accommodate changing requirements in existing architecture. We investigate change logs in order to analyse operational representation of architecture change instances to discover an implicit evolution-centric knowledge that have been aggregating over time. Change instances from the log are formalised as a typed attributed graph with its node and edge attribution capturing change representation on architecture elements. We exploit graph matching as a knowledge discovery technique in order to i) analyse change operationalisation and its dependencies for ii) discovering recurrent change sequences in the log. We identify potentially reusable, usage-determined change patterns.

A framework for acquisition and application of software architecture evolution knowledge

“…In Figure 1, an architecture evolution history is represented as a source of knowledge that consists of traces of architecture-centric changes maintained during evolution of software architectures. Knowledge Source represents a transparent and centrally manageable repository [11,12] as a recorded collection of architecture change representations [9,13]. It provides a foundation and fine-grained change representations for experimental analysis of real data concerning architecture evolution.…”

“…In software engineering and software evolution literature, the terms modernisation and evolution are virtually synonymous and often used interchangeably -referring to architecture-based change implementation [1,13]. In this section, we explain some existing frameworks that are used as reference models to guide the architecture evolution process.…”

“…PatEvol only considers reuse of evolution-centric knowledge that is not addressed in any of the existing frameworks. More specifically, in the context of architecture change analysis and management, the PatEvol framework lies at the intersection of two distinct research areas: i) software repository mining [11,12] to discover evolution knowledge from architecture evolution histories [13] and ii) software evolution [1,2] that relies on the discovered knowledge to support architectural evolution [10]. In conceptual terms, the PatEvol framework adapts the basic ideas from the SOA migration horseshoe [15].…”

“…More specifically, the knowledge source or architecture change logs [9,13] represent a central repository that contains fine-grained instances of architecture change and provides a foundation for evolution mining. We propose a catalogue of architecture evolution patterns that promotes an empirically discovered collection of patterns as reusable solutions to recurring problems of software architecture evolution.…”

“…Our objective for evolution mining is identical to that of software evolution analysis [12,11], which exploits the history of a software system to analyse its present state and to predict its future. In the context of software repository mining, architecture evolution mining is aimed at employing a set of (automated) techniques for extraction of architecture change instances from change logs [9,13]. Therefore, we exploit architecture change logs that provide fine-grained details about architecture change instances.…”