We posit that large-scale conservative simulations showing both discrete-event and agent-based characteristics will greatly benefit from a unified, distributed-memory interaction paradigm based on the space-time memory (STM) abstraction. We will build a performant C/C++ library which implements an STM data-structure, and a minimal interface sufficient to support the development of a conservative parallel discrete event simulation (PDES) engine by leveraging the remote memory access (RMA) functionality of the latest message passing interface (MPI) protocol. We will build an extremely scalable conservative PDES engine on top of the STM library in a scripting language (Python/Lua) – the relevant methods of the STM library will be exposed using foreign function interface mechanism for shared libraries. We will analyze bottlenecks in our architecture by implementing specific micro-benchmarks using the LA-PDES benchmark suite.

Multi-agent models coupling AI and non-AI agents using STM-PDES, enables simulations at scale allowing for rule-based controlled divergence, multi-objective optimization and radical emergent responses at runtime. We will perform several medium to large-scale simulations comparing the performance of STM-PDES AI-agent models to the more conventional PDES engine-based AI-agent simulations. We will make available all software and data generated as part of this project as open source to the broader research community.