Balanced and Efficient Distribution of Coarse and Nested Grids in Regional Ocean Modeling System

Regional Ocean Modeling System (ROMS) applications often employ nested grid configurations to capture fine-scale processes more accurately. In this study, we plan to accelerate nested grid simulations for the Bay of Bengal (BoB) region using ROMS. We successfully implemented a 2.3km resolution application for the BoB region, which served as the foundation for setting up a Refinement-type nested grid application. Our analysis revealed that the default strategy, which distributes all grids equally across processors, was inefficient for the refinement grids. These grids typically require fewer computational resources than coarser grids, resulting in increased communication latency and limitations in scalability. We propose a new workload distribution strategy which solves the problem of scaling nested grid applications in ROMS, specifically for the nested grids. Moving forward, we plan to develop an improved distribution strategy that optimizes both workload and inter-processor communication. Additionally, we aim to extend this strategy to various superclasses of nested grids supported by ROMS, ensuring efficient execution across different scenarios.