CHAMB-GA: Containerized framework scales genetic algorithms to 3,500+ cores

CHAMB-GA is a new open-source framework from researchers at RWTH Aachen University and Forschungszentrum Jülich, designed to streamline metaheuristic global optimization that involves long-running embedded simulations. The framework uses a microservice architecture where users provide GA operators and simulation backends separately. A central message broker (e.g., RabbitMQ) coordinates asynchronous manager-worker communication, enabling parallelized evolution and fitness evaluation across distributed hardware. This decoupling allows different parts of the compute infrastructure to handle distinct tasks, overcoming the rigidity of traditional monolithic optimization tools.

In benchmarks, CHAMB-GA scaled to over 3,500 CPU cores with only minimal overhead, demonstrating near-linear scaling on a classical optimization test function. The framework also succeeded in a real-world case: optimizing the dispatch of High Voltage Direct Current (HVDC) lines in the German transmission grid. It seamlessly migrated from a local Kubernetes deployment to a large SLURM-based HPC cluster, combined horizontal and vertical scaling, and integrated multi-stage workflows. The containerized approach ensures portability and reproducibility across environments, from personal computers to exascale systems.