Efficient simulation of agent-based models on multi-GPU and multi-core clusters

Abstract

An effective latency-hiding mechanism is presented in the parallelization of agent-based model simulations (ABMS) with millions of agents. The mechanism is designed to accommodate the hierarchical organization as well as heterogeneity of current state-of-the-art parallel computing platforms. We use it to explore the computation vs. communication trade-off continuum available with the deep computational and memory hierarchies of extant platforms and present a novel analytical model of the tradeoff. We describe our implementation and report preliminary performance results on two distinct parallel platforms suitable for ABMS: CUDA threads on multiple, networked graphical processing units (GPUs), and pthreads on multi-core processors. Message Passing Interface (MPI) is used for inter-GPU as well as inter-socket communication on a cluster of multiple GPUs and multi-core processors. Results indicate the benefits of our latency-hiding scheme, delivering as much as over 100-fold improvement in runtime for certain benchmark ABMS application scenarios with several million agents. This speed improvement is obtained on our system that is already two to three orders of magnitude faster on one GPU than an equivalent CPU-based execution in a popular simulator in Java. Thus, the overall execution of our current work is over four orders of magnitude faster when executed on multiple GPUs.

Multi-agent simulation on multiple GPUs

“…Nevertheless, [27] showed that OpenCL performs worse than CUDA. The work in [28] that describes latency hiding schemes for cellular-based models is one of the few existing related work for solving the scaling issue. These together with the need for better visualization and GUI tools for real-time analysis motivated the work in this paper.…”

A Parallel Multi-agent Spatial Simulation Environment for Cluster Systems

“…Although many parallelization attempts have been made to multi-agent simulations, most parallelization work has been limited to shared-memory programming environments such as multithreading, OpenMP, and CUDA [2], [3], [4], [5]. It is much more challenging to parallelize multi-agent simulation on distributed-memory systems.…”

Multi Agent System Optimization in Virtual Vehicle Testbeds

“…These approaches, and other evaluation of GPU based MAS such as [3], show that GPU based MAS can incorporate a drastic performance improvement up to an average factor of 60 with respect to purely CPU based implementations.…”