Waffle: A Flexible Framework for Large-scale Social Simulation Experiment

Abstract

We propose Waffle, a novel experiment workflow control framework for large-scale social simulation experiments utilizing high-performance computing clusters (HPCCs). Advancement in computer technology has enabled large-scale experiments in computational social science. However, with the expansion in the scale of the experiment, the cost of manually managing the experiment increases considerably. To address this issue, we provide the experiment workflow control framework that uses a recursive meta-level control mechanism and a chain of procedures by the guarded suspension to automate management. The framework manages task queuing in parallel processing, experimental results, and HPCC jobs. We evaluate the functionality and efficiency of the proposed framework using a case study, which is a data assimilation experiment involving virus infection simulation. The framework supports evidence-based policymaking against long-term urban planning or emergencies, such as pandemics, because of the ability to handle large-scale experiments.