TeraAgent: A Distributed Agent-Based Simulation Engine for Simulating Half a Trillion Agents Conference: ; ; CCS: Computing methodologies Massively parallel and high-performance simulations CCS: Computing methodologies Agent / discrete models CCS: Computing methodologies Distributed algorithms CCS:

Lukas Johannes Breitwieser, Ahmad Hesam, Abdullah Giray Yağlıkçı, Mohammad Sadrosadati, Fons Rademakers, Onur Mutlu · 2025

Read TeraAgent: A Distributed Agent-Based Simulation Engine for Simulating Half a Trillion Agents Conference: ; ; CCS: Computing methodologies Massively parallel and high-performance simulations CCS: Computing methodologies Agent / discrete models CCS: Computing methodologies Distributed algorithms CCS: by Lukas Johannes Breitwieser, Ahmad Hesam, Abdullah Giray Yağlıkçı, Mohammad Sadrosadati, Fons Rademakers, Onur Mutlu (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2509-24063.txt.

Introduction

Simulating a living organ cell by cell can need billions of small agents that act and react. Lukas Breitwieser and 5 co-authors built TeraAgent, an engine that spreads such simulations across many servers. The prior leading tool, BioDynaMo, ran on one server only, which capped its size.

TeraAgent packs agent data so it can be read and changed right in the receive buffer. It also sends only what changed between steps, which cuts traffic a lot. The engine ran a simulation with half a trillion agents on tens of thousands of cores. It packed data far faster than a standard baseline and sped up live viewing many times over.

Catalog record

CREATORLukas Johannes Breitwieser, Ahmad Hesam, Abdullah Giray Yağlıkçı, Mohammad Sadrosadati, Fons Rademakers, Onur Mutlu
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE54 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-24063

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

TeraAgent: A Distributed Agent-Based Simulation Engine for Simulating Half a Trillion Agents Conference: ; ; CCS: Computing methodologies Massively parallel and high-performance simulations CCS: Computing methodologies Agent / discrete models CCS: Computing methodologies Distributed algorithms CCS: Software and its engineering Software performance Lukas Johannes Breitwieser, Ahmad Hesam, Abdullah Giray Yağlıkçı, Mohammad Sadrosadati, Fons Rademakers, Onur Mutlu Abstract . Agent-based simulation is an indispensable paradigm for studying complex systems. These systems cancomprise billions of agents, requiring the computing resources of multiple servers to simulate. Unfortunately, the state-of-the-art platform, BioDynaMo, does not scale out across servers due to its shared-memory-based implementation. To overcome this key limitation, we introduce TeraAgent, a distributed agent-based simulation engine. A critical challenge in distributed execution is the exchange of agent information across servers, which we identify as a major performance bottleneck. We propose two solutions: 1) a tailored serialization mechanism that allows agents to be accessed and mutated directly from the receive buffer, and 2) leveraging the iterative nature of agent-based simulations to reduce data transfer with delta encoding. Built on our solutions, TeraAgent enables extreme-scale simulations with half a trillion agents (an 84× improvement), reduces time-to-result with additional compute nodes, improves interoperability with third-party tools, and provides users with more hardware flexibility. obeypunctuation=true] ETH Zurich , , Switzerland obeypunctuation=true] Delft University of Technology , , The Netherlands obeypunctuation=true] ETH Zurich , , Switzerland obeypunctuation=true] ETH Zurich , , Switzerland obeypunctuation=true] CERN , , Switzerland obeypunctuation=true] ETH Zurich , , Switzerland 1. Introduction Agent-based modeling is a bottom-up simulation method to study complex systems. Early agent-based models, dating back as far as 1971, studied segregation in cities ( Schelling 1971 ) , flocks of birds ( Reynolds 1987 ) , and complex social phenomena ( Epstein and Axtell 1996 ) . These models follow the same three-step structure, although the studied systems differ widely. First, the researcher has to define what the abstract term agent should represent. In the flock model from Craig Reynolds ( Reynolds 1987 ) , an agent represents a bird. The bird has a current position, and a velocity vector. Second, the researcher has to define the agent’s behaviors. For example, birds avoid collision, align their velocity with neighbors, and try to stay close to each other. Third, the researcher has to define the starting condition of the simulation. How should the birds be distributed in space in the beginning? What are their values for the initial velocity, and parameter values for the three behaviors (collision avoidance, velocity matching, and flock centering)? Once all these decisions are made, the model is given to the simulation engine, which executes it for a number of iterations, or until a specific condition is met. Agent-based models exhibit two characteristics: local interaction and emergent behavior. First, agents only interact with their local environment and do not have knowledge about agents that are “far” away. This characteristic makes a distributed execution feasible via spatial partitioning of the agents, which results in data exchanges across partitions at the bordering regions. Second, agent-based models show emergent behavior or in other words they demonstrate that: “the whole is more than the sum of its parts”. To come back to the flock of birds example, the swarm dynamics were not programmed into the model, but arose solely through the interactions of the birds following the three behaviors. Since these early studies ( Schelling 1971 ; Reynolds 1987 ; Epstein and Axtell 1996 ) agent-based models have been used in numerous domains including biology ( de Montigny et al. 2021 ; Duswald et al. 2024b ; Cogno et al. 2024 ; Cogno 2023 ; Cogno et al. 2022b ; Cogno et al. 2022a ; Nicolò Cogno and Vavourakis 2024 ; Metzcar et al. 2019 ; Wang et al. 2015b ; Olsen and Siegelmann 2013 ; Wang et al. 2013 ; An and Kulkarni 2015 ; Wang et al. 2015a ; Wang et al. 2015c ; Poleszczuk et al. 2016 ; Norton et al. 2019 ; Heidary et al. 2020 ; Rivera et al. 2022 ; van Genderen et al. 2024 ; Zubler and Douglas 2009 ; Duswald et al. 2024a ; Torben-Nielsen and De Schutter 2014 ; De Schutter 2023 ; Abubacar and Bauer 2024 ; Zubler 2009 ; Pennisi et al. 2015 ; Caffrey et al. 2014 ; Avin et al. 2021 ; Glen et al. 2019 ; Tang et al. 2011 ; Lambert et al. 2018 ; Walpole et al. 2015 ; Camacho-Gómez et al. 2022 ; Dalle Nogare and Chitnis 2020 ; Bonabeau 1997 ; Thorne et al. 2007 ; Kreft et al. 2001 ; Wilmoth et al. 2018 ; Li et al. 2024 ; Latif and May 2018 ; Sweeney et al. 2019 ; Nagarajan et al. 2022 ; Koshy-Chenthittayil et al. 2021 ) , medicine ( Cogno et al. 2024 ; Cogno 2023 ; Cogno et al. 2022b ; Cogno et al. 2022a ; Montealegre and Rammig 2012 ; Shinde and Kurhekar 2018 ; Bauer et al. 2009 ; Truszkowska et al. 2022 ; Perrin et al. 2006 ; Pappalardo et al. 2018 ; Folcik et al. 2011 ; Folcik et al. 2007 ; Chiacchio et al. 2014 ; Jacob et al. 2011 ; Li 2016 ; Montagna and Omicini 2017 ; Nianogo and Arah 2015 ; Schryver et al. 2015 ; Mi et al. 2007 ; Khademi et al. 2018 ; Li et al. 2014 ; Archbold et al. 2024 ; Giabbanelli et al. 2021 ; Morshed et al. 2019 ; Beheshti et al. 2017 ; Giabbanelli and Crutzen 2017 ; Squires et al. 2023 ; Gao et al. 2017 ; Li et al. 2018 ) , epidemiology ( Hesam et al. 2024 ; Ozik et al. 2021 ; Yin et al. 2021 ; Kou et al. 2021 ; Faucher et al. 2022 ; Pescarmona et al. 2021 ; Cattaneo et al. 2022 ; Kerr et al. 2021 ; Truszkowska et al. 2023 ; Hoertel et al. 2020 ; Kumaresan et al. 2023 ; Ogden et al. 2024 ; Topîrceanu 2023 ; Laskowski et al. 2011 ; DePasse et al. 2017 ; Karimi et al. 2015 ; Kumar et al. 2013 ; Teweldemedhin et al. 2004 ; Brookmeyer et al. 2014 ; Anderle et al. 2024 ; Woodul 2023 ; Venkatramanan et al. 2018 ; Tabataba et al. 2017 ; Macal et al. 2018 ; Hackl and Dubernet 2019 ; Martono and Ohwada 2024 ; Buttenheim et al. 2013 ; Hunter et al. 2017 ; Miksch et al. 2019 ; Estill et al. 2021 ; Tracy et al. 2018 ) , finance ( Axtell and Farmer 2024 ; Palmer et al. 1994 ) , policy making ( Farmer and Foley 2009 ) and many more ( Macal and North 2014 ; Huber et al. 2018 ; Boone 2017 ; Castro et al. 2020 ; Stieler et al. 2022 ; Heppenstall et al. 2021 ; Bauer et al. 2009 ; North et al. 2010 ; Romero et al. 2023 ; Schramm et al. 2010 ; Negahban and Yilmaz 2014 ; Rand and Stummer 2021 ; Garcia et al. 2007 ; Rand and Rust 2011 ; Tian and Chang 2020 ; Robinson and Rai 2015 ; Sachs et al.

Read the complete text