Continuous Integration and Continuous Deployment (CI/CD) pipelines are crucial for modern software development, providing rapid feedback and enabling automated delivery. While numerous patterns, best practices, and anti-patterns have been proposed to improve pipeline performance, their actual impact on execution time can vary considerably across projects. Currently, practitioners lack predictive methods to estimate how structural changes to a pipeline affect its overall performance, often resorting to trial and error. To address this issue, we present a performance optimization model for CI/CD pipelines that considers job dependencies and predicts the impact of structural refactorings on overall execution time. By representing pipelines as directed acyclic graphs (DAGs) and modeling job execution through a multi-stage priority queuing approach, our method captures both job dependencies and pipeline overheads. We formalize common refactorings (including parallelization, sequentialization, job splitting/merging, and environment-specific adjustments) and provide constraints to ensure correctness and feasibility. We evaluate our approach on a combination of synthetic pipelines, open-source Software Engineering projects, and Reinforcement Learning pipelines designed to reflect diverse training and evaluation workflows. Our results demonstrate that the model accurately predicts changes in pipeline duration, offering practitioners actionable insights to select the most effective performance-enhancing strategies. This work, to the best of our knowledge, is the first to provide a holistic, predictive framework for estimating the performance impact of structural refactorings in CI/CD pipelines, thereby bridging the gap between empirical best practices and quantitative optimization.

Top