Software development in industry is getting increasingly complex as systems are getting more sophisticated and are often interconnected constituting the system landscape. Architecture description is therefore getting increasingly important. The necessary maintenance of description is often neglected because of diferent priorities due to time and budget constraints. This leads, among other things, to outdated architecture description. For a more eficient planning of the architectural landscape and prevention of redundancy, it is vital that architects and other stakeholders have the most current information about the systems. This paper presents doctoral research in its early stages concerned with the issue of continuous architecture recovery allowing to reflect the current architecture and evolution of the system as a digital architecture twin. The proposed approach aims to automatically extract architecture information of complex systems by recovering it from heterogeneous architectural data sources. The idea is the consolidation and integration of this recovered architecture information at diferent points in time to enable the representation of the systemand its evolution. This permits the use of an architecture information query language facilitating diferent use cases (e.g., support of architectural design decisions or tailored architecture description). Planned contributions are the assessment and consolidation of heterogenous information sources and the application of architecture recovery methods with the noteworthy addition of versions over time of those information sources and the creation of a co-evolving digital twin.
One of the main problems in software architecture evolution and maintenance is the low
quality and even non-existence of architecture description (e.g., architecture models) as systems
evolve and increase in complexity, and are adapted to environments, technology or customer
requirements. Evolution of a system should entail evolution and maintenance of its description,
as otherwise the description does not reflect the actual system anymore, resulting in a decrease
of quality and usefulness of the architecture description. Yet, the creation and maintenance of
architecture description is linked with high efort (time and costs) as it is a primarily manual
task. However, an updated description is a key driver for an architect to understand a system,
comprehend dependencies and decide on future enhancements. Furthermore, stakeholders
require diferent views [
Automated architecture recovery approaches range from static to dynamic methods using
diferent techniques like structural clustering, concern-based clustering, or interactive exploration to
extract and recover diferent layered architecture information and employing input parameters
like the implemented system (e.g., as source code). Approaches like ACDC [
The idea of the approach is to automatically create and co-evolve a Digital Architecture Twin
(DArT) of heterogeneous and evolving systems. In general, a Digital Twin is a virtual
representation of a physical or non-physical object (Physical Twin) or process often used in the
digitalization of cars or engines and enabling the exchange of data and information between
the Digital Twin and Physical Twin. This allows to efectively simulate situation and adaptions
without tinkering with the real world image, which could result in high costs or unfavorable
failures [
Co-Evolution Architecture information Virtual representation
Coordination of heterogeneous Architecture Information data sources Recovery Services
Incremental updates of the Digital Twin Architecture information sources System
Data Collection
Agents
Integration & Consolidation module
Versioning of the Digital Architecture Twin mUecphdaanteis-m consRoelidcaotveedreindfoarnmdation
Digital Architecture
Twin
Architecture query language Persistence of the Digital Twin
Recommendation system for architectural design decisions
Pattern and Style Discovery
Tailored Documentation
Automated continuous compliance checking and quality assessment
The Digital Architecture Twin generation process, shown in figure 1, begins with the collection of architectural data from architecture information sources using Data Collection Agents (DCA). Next is the provisioning of this data for the architecture recovery methods implemented as Architecture Information Recovery Services (AIRS). For this, we combine proven architecture information recovery approaches leveraging diferent data sources. We integrate and consolidate the results provided by the architecture recovery methods into a unified architecture information model based on meta models representing the Digital Architecture Twin to obtain an overarching representation of the system architecture. When the system evolves (e.g., source code changes), the DCA and AIRS update the existing architecture information maintaining information of old versions. For this, the update-mechanism triggers the AIRS either automatically or periodically, depending on the source, to retrieve the current architecture information. This updates the DArT incrementally and keeps it up to date with the evolving system. An open challenge is to ensure that the DArT is conform with the system. To use the collected architecture information, an architecture query language will be developed to dynamically retrieve architecture information of diferent versions, views or abstractions levels of a system and its architecture information.
The DArT in combination with the architecture query language enables to dynamically query information that can be tailored to the specific requirements of developers, architects and other stakeholders at the desired abstraction level and system version. An optional visualization of the dynamic queries and views shall result in human-readable architecture description.
Potential use cases of the DArT are: guided architecture design via architecture recommendation and continuous compliance checking to prevent architecture drift. Guided architecture designs allow based on specific questions and similarity matching of existing architecture information tailored architectural proposals enabling a consolidated IT landscape. Automated and continuous compliance checking monitors whether the actual system has diverged from the planned design (software architecture drift/erosion). Recovered architecture information (as-is) can be compared to the explicitly documented system architecture (as-planned) in order to detect and counteract increased erosion at an early stage.
The process of the doctoral research is displayed in Fig. 2. The first step is a systematic literature review concerned with architecture recovery methods to get a thorough overview of existing approaches, their potential use cases and benefits as well as their limitations. The second step is the identification of potential data sources for the extraction of architecture data, which can be used to recover architecture information employing architecture recovery methods. The third step is the implementation of suitable architecture recovery methods. The fourth step is the creation of the DArT by consolidating the recovered architecture information into an architecture information model built specifically for the integration of static, dynamic and deployment information. The fifth step is the development of an architecture query language built for the retrieval of tailored, stakeholder-dependent architecture information employing the DArT. The sixth step is the conduction of case studies and expert interviews, which are performed iteratively, to evaluate the benefits and understand possible customization of the approach. This evaluation is done using a prototype, which will be developed to extract the architecture and evolution information which enables the generation of the DArT.
Research question: "How to automatically generate the architecture description of an evolving system?" SLR
Identification of data sources
Implementation of DCA and AIRS
Creation of
Digital Architecture Twin
Architecture query language
Case studies/ Evaluation
This work contributes the following: (1) development of the conceptual idea of the DArT for the description of a system, (2) consolidation of available and recovered architecture information in a DArT, (3) co-evolution of the DArT with the system using incremental updates featuring heterogeneous architecture artifacts from diferent points in time and (4) development of a query language capable of retrieving architecture information for tailored stakeholder perspectives using the DArT.
We plan to evaluate our approach by developing a prototype of the proposed approach, applying it to open source applications and real-world applications in industry. Furthermore, we intend to perform expert interviews to gather feedback on the idea and approach. Possible limitations of the approach can be the resource- and time intensive architecture recovery process, leading to time shifted description of the analyzed system, devaluing the digital twin. Another limitation is the dificult integration of heterogeneous sources and the consolidation of potentially contradicting information extracted from diferent sources. Future work comprises of the development of the exchange from DArT to system.