EP-SOA: A Framework to Enable Transition of High-Fidelity Codes to Heterogeneous Computing Environments

DTRA uses HF codes to investigate weapon effects phenomenology and techniques for Countering Weapons of Mass Destruction (CWMD). End-to-end High-Fidelity simulations in support of the DTRA Agent Defeat Modeling Baseline require calculations across multiple phenomena occurring at vastly different timescales, from microseconds to hours. As DTRA increasingly relies on computational modeling and simulation tasks executed on HPC systems and as HPCMP shifts towards deploying more performant heterogeneous systems, there is a pressing need to modernize and transition the simulation codes.Refactoring legacy HF codes to target heterogeneous architectures, however, is very challenging. HF codebases are large, often consisting of hundreds of thousands of lines of code. The codes are written in traditional programming paradigms and have strict testing requirements to ensure simulation results remain highly accurate. Porting the entire codebase to a heterogeneous environment is, therefore, not practical. Developers need robust tools and methodologies to identify which computational hotspots to offload to accelerators and which to retain on CPUs and understand the necessary code changes for optimal accelerator performance. EP Analytics proposes to develop EP-SOA (for EP Analytics' Smart Offload Assistant) to fill this need for robust tools. EP-SOA will facilitate an incremental, targeted, and manageable transition to heterogeneous systems, thereby vastly reducing the efforts required to adapt the codes.The result of the EP-SOA-driven transition to heterogeneous environments will directly impact DTRA’s mission. DTRA’s simulations often run for several months, consuming substantial HPC resources. Effectively utilizing heterogeneous environments can considerably accelerate these simulations, allowing for completion well ahead of deadlines. Such acceleration, combined with the potential to run more comprehensive simulation ensembles, provides decision-makers with a deeper understanding of the simulated scenarios.