Mastering Reference Architectures with Modeling Assistants

Reference architectures (RA) can significantly simplify and speed up the design and evolution of software systems in virtually any domain. Rather than starting from scratch or trying to reuse models not meant for reuse, software architects can exploit RAs to initiate their designs from rigorously-defined architectural models specifically defined to support reuse. RAs can guarantee the satisfaction of key functional and non-functional requirements, including those related to technical and societal challenges, such as security and sustainability, significantly reducing the design effort and improving the design quality. The growing popularity of RAs is confirmed by the many providers, including Amazon AWS, Microsoft Azure, IBM, and Salesforce, that offer RAs for application deployment on their infrastructures. Unfortunately, RAs are expensive to define since they must comprehensively synthesize the design knowledge of specific application domains (e.g., e-commerce, automotive, etc.) and technical contexts (e.g., cloud systems, mobile apps, IoT, etc.). This challenge limits the availability of RAs to a few domains and contexts, hindering their adoption. We envision the opportunity to facilitate RA-based design by introducing a novel combination of mining in-the-wild and continuous knowledge refinement through a framework named ASSISTRA. Mining in-the-wild consists of bots crawling architectural models at scale (e.g., from public repos) and analyzing them continuously to automatically synthesize RAs and extract RAs-related practices that shall feed knowledge bases. Continuous knowledge refinement consists of a population of recommender systems that, while interactively supporting the architects in their activity using the shared knowledge bases, shall continuously refine the mined content in the knowledge base (e.g., based on architects accepting, modifying, or rejecting suggestions). This process shall generate high-quality recommendations, drastically simplifying the design activity. This paper elaborates on this vision, highlights the key challenges, and calls for novel research contributions.