A MULTI‑OBJECTIVE OPTIMIZATION MODEL FOR DESIGNING THE SOFTWARE ARCHITECTURE OF INTERNET‑OF‑THINGS SYSTEMS

Abstract

Modern Internet of Things (IoT) systems span edge, fog, and cloud layers and must satisfy several conflicting non-functional requirements, such as responsiveness, security, reliability, and resource efficiency. Earlier work (CAL and MCAL) increased the transparency of architectural decisions by scoring and ranking individual patterns using a weighted sum, but implicitly treated the architecture as a loose set of independent patterns and compressed all requirements into a single scalar index. In this paper, IoT architecture design is reformulated as a multi-objective pattern-portfolio selection problem. Each candidate architecture is represented as a binary portfolio of patterns and evaluated by an additive vector of objective functions corresponding to selected quality attributes, while linear constraints capture practical engineering limits: implementation budget, pattern dependencies, incompatibilities, and portfolio size. The resulting multi-objective integer linear programming model yields a discrete Pareto set of non-dominated portfolios that makes trade-offs explicit and auditable. A small case study with two antagonistic objectives, responsiveness and security, illustrates the effect of the reformulation and compares it with the classic weighted-sum approach. The scalar model produces markedly different "optimal" portfolios for different weight vectors and can hide feasible compromise solutions that balance competing requirements. In contrast, the Pareto-based model exposes the entire family of efficient portfolios under the same data and constraints, allowing architects and stakeholders to select an option that matches their priorities, constraints, and acceptable risk, providing a clear basis for subsequent a-posteriori ranking or sensitivity analysis when project preferences change. The formulation is readily extensible to additional objectives and larger pattern catalogs, keeping feasibility rules explicit, and reusable across IoT projects, thus supporting systematic architectural decision-making in complex distributed environments.