WS-Line: Workload scheduling through LLM-assisted DSL generation and CSOP modeling

Context: Workload Scheduling in Service-Oriented Architectures requires coordinating services and tasks under strict temporal and resource constraints. Constraint-based formulations, particularly Constraint Satisfaction Optimization Problems (CSOPs), provide formal guarantees for computing feasible and optimized workload schedules. However, adapting Source Requirements (SR) to formal workload scheduling models remains a manual, time-consuming, and error-prone process that requires significant domain expertise. Objective: This paper aims to automate the transformation of Source Requirements into a formally analyzable Scheduled Execution Manifest, while preserving the rigor and optimization guarantees of constraint-based workload scheduling. Methods: We present WS-Line, an end-to-end approach that integrates Large Language Models (LLMs), Retrieval-Augmented Generation (RAG), a formally defined Domain-Specific Language (WorkSched-DSL), and a CSOP-based optimization model specifically designed for workload scheduling. The approach extracts structured specifications from Source Requirements documentation, validates them against a predefined DSL schema, translates them into data representations, and computes optimized schedules through constraint programming, obtaining correct Scheduled Execution Manifests (SEMs). Results: The evaluation, conducted on real-world aerospace use cases and progressively scaled synthetic workloads, assesses (i) the computational performance of the CSOP model, (ii) the syntactic and semantic correctness of the generated DSL specifications, and (iii) the consistency of the final Scheduled Execution Manifests. Results show consistent syntactic validity, high semantic fidelity, and efficient schedule computation. Conclusion: WS-Line bridges the gap between Source Requirements and formally verified workload scheduling models. By combining LLM- and RAG-based information extraction with constraint-based optimization, the approach enables automation of workload scheduling while maintaining correctness, modularity, and extensibility in service-oriented environments.