Repast HPC

Repast HPC (Repast for High Performance Computing) is a system for agent-based models, simulations of many interacting individuals, on large-scale distributed computing platforms.^1,2 One model runs on many processors at once, and they exchange data through the MPI message-passing standard.^2,6 The toolkit is written in C++ and runs on macOS, Linux and Unix; models can be written in standard C++ or in a Logo-style C++.² It is lean and aimed at experts.^1,3

Repast HPC carries the core ideas of Repast Simphony, the suite's Java toolkit, over to distributed computing.^1,2 It has been tested for scalability on machines from the Top500 list of supercomputers, including Argonne National Laboratory's Mira and Theta.² Building it needs the Boost, NetCDF and curl libraries and an MPI installation; the download is a tar archive with installation instructions, and there is a tutorial and a manual.^3,6

The current version, 2.3.1, was released on 21 October 2021, and there has been no release since.^1,7 The suite is copyright Argonne National Laboratory and released under a New BSD style licence.⁵ The paper to cite is Collier and North, "Parallel Agent-Based Simulation with Repast for High Performance Computing", SIMULATION, 2013.⁴

The Repast Suite has two other members.¹ Repast Simphony is a Java toolkit for workstations and small clusters, described as interactive and easy to learn; version 2.12.0 came out on 23 October 2026.¹ Repast4Py, version 1.3.1 from 22 October 2026, is a Python toolkit for distributed agent-based models, intended as an easier way in for researchers who want large-scale models.¹

Repast HPC is for a model of a place that has outgrown one computer, such as one with an agent for every resident of a region, and for a team that can write and build C++ on a cluster. For a first model, Repast Simphony, Repast4Py, Mesa or NetLogo are easier starts; FLAME GPU reaches large populations on graphics cards instead.