Distributed Memory Parallel Implementation of Agent-Based Economic Models
We present a Distributed Memory Parallel (DMP) implementation of agent-based economic models, which facilitates large-scale simulations with millions of agents. A major obstacle in scalable DMP implementation is to distribute a balanced workload among MPI processes, while making all the topological graphs, over which the agents interact, available at a minimum communication cost. We addressed this problem by partitioning a representative employer-employee interaction graph, and all the other interaction graphs are made available at negligible communication costs by mimicking the organizations of the real-world economic entities. Cache-friendly and low-memory intensive algorithms and data structures are proposed to improve runtime and scalability, and the effectiveness of each is demonstrated. The current implementation is capable of simulating 1:1 scale models of medium-sized countries.
KeywordsAgent-based economic models Large-scale simulations MPI
This work was supported by JSPS kakenhi grant 18H01675. Parts of the results are obtained using K computer at the RIKEN Center for Computational Science, and the Reedbush supercomputer at the Univ. of Tokyo.
- 6.Dawid, H., Delli Gatti, D.: Agent-based macroeconomics. In: Hommes, C., LeBaron, B. (eds.) Handbook of Computational Economics, vol. 4, pp. 63–156 (2018)Google Scholar
- 7.Poledna, S., et al.: When does a disaster become a systematic event? Estimating indirect economic losses from natural disasters. arXiv:1801.09740
- 9.Balaji, P., Chan, A., Gropp, W., Thakur, R., Lusk, E.: Non-data-communication overheads in MPI: analysis on Blue Gene/P. In: Lastovetsky, A., Kechadi, T., Dongarra, J. (eds.) EuroPVM/MPI 2008. LNCS, vol. 5205, pp. 13–22. Springer, Heidelberg (2008). https://doi.org/10.1007/978-3-540-87475-1_9CrossRefGoogle Scholar