A Survey on Agent-based Simulation Using Hardware Accelerators

Xiao, Jiajian; Andelfinger, Philipp; Eckhoff, David; Cai, Wentong; Knoll, Alois

doi:10.1145/3291048

Cited by 38 publications

(26 citation statements)

References 161 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To implement the two-state access graph, two sets of state variables are used, 1 and 2 . In contrast, the independent and ordered updates 4, agent or can be processed at the same time, e.g., making the update scheme suitable for execution on parallel computing platforms, e.g., GPUs or FPGAs [39,40].…”

Section: Cyclic Dependent Agentsmentioning

confidence: 99%

Causality and Consistency of State Update Schemes in Synchronous Agent-based Simulations

Tan

Andelfinger

Eckhoff³

et al. 2021

Proceedings of the 2021 ACM SIGSIM Conference on Principles of Advanced Discrete Simulation

Self Cite

View full text Add to dashboard Cite

In an agent-based simulation (ABS), a state update scheme carries out the transitions of agents from one state to the next. To produce correct simulation results, the update scheme must respect the cause-and-effect relationships defined by the agent-based model and ensure that the resulting overall simulation state is internally consistent. At the same time, the update scheme should be efficient enough to meet a simulationist's demand for timely results. Considering the common class of synchronous time-driven ABS, a number of update schemes have been employed in the literature and simulation frameworks. In this paper, various implementations of update schemes are analyzed and contrasted with respect to their ability to maintain the simulation correctness as well as their performance characteristics. A semantic model is formulated to define the reference behavior of synchronous time-driven ABS updates and model the dependencies among agent updates using a state access graph. Relying on the formalization, conditions under which different update schemes achieve causality are shown. Further, resolution methods are categorized according to their coordination mechanisms to achieve consistency by resolving conflicts among agent state updates. Through two case studies, the empirical performance of different update schemes and resolution methods are evaluated. For sequential execution, an update scheme based on the agent's dependencies achieves the highest performance, whereas in the parallel case, the choice of update scheme involves a tradeoff between execution time and memory usage. If deterministic simulation output is required, decentralized coordination generally outperforms centralized coordination. The results can assist implementers and researchers in their selection of appropriate methods in the design and implementation of agent-based simulators.

show abstract

Section: Cyclic Dependent Agentsmentioning

confidence: 99%

Causality and Consistency of State Update Schemes in Synchronous Agent-based Simulations

Tan

Andelfinger

Eckhoff³

et al. 2021

Proceedings of the 2021 ACM SIGSIM Conference on Principles of Advanced Discrete Simulation

Self Cite

View full text Add to dashboard Cite

show abstract

“…Some recent research aims at integrating differentiable programming facilities into machine learning frameworks such as Py-Torch [41]. Implementing differentiable agent-based models within such frameworks would enable an efficient unification of simulationbased optimization and neural network training, making use of the frameworks' optimized GPU-based implementations of neural networks and automatic differentiation while accelerating the model execution through fine-grained many-core parallelism [64,65].…”

Section: Limitations and Research Directionsmentioning

confidence: 99%

Differentiable Agent-Based Simulation for Gradient-Guided Simulation-Based Optimization

Andelfinger

2021

Proceedings of the 2021 ACM SIGSIM Conference on Principles of Advanced Discrete Simulation

Self Cite

View full text Add to dashboard Cite

Simulation-based optimization using agent-based models is typically carried out under the assumption that the gradient describing the sensitivity of the simulation output to the input cannot be evaluated directly. To still apply gradient-based optimization methods, which efficiently steer the optimization towards a local optimum, gradient estimation methods can be employed. However, many simulation runs are needed to obtain accurate estimates if the input dimension is large. Automatic differentiation (AD) is a family of techniques to compute gradients of general programs directly. Here, we explore the use of AD in the context of time-driven agent-based simulations. By substituting common discrete model elements such as conditional branching with smooth approximations, we obtain gradient information across discontinuities in the model logic. On the example of microscopic traffic models and an epidemics model, we study the fidelity and overhead of the differentiable models, as well as the convergence speed and solution quality achieved by gradient-based optimization compared to gradient-free methods. In traffic signal timing optimization problems with high input dimension, the gradient-based methods exhibit substantially superior performance. Finally, we demonstrate that the approach enables gradient-based training of neural network-controlled simulation entities embedded in the model logic. CCS CONCEPTS• Mathematics of computing → Automatic differentiation; • Theory of computation → Continuous optimization; Multiagent reinforcement learning; • Computing methodologies → Modeling methodologies; Agent / discrete models.

show abstract

“…A wide variety of dedicated hardware accelerators have been developed to accelerate a wide range of general computing functions, e.g., simulations [208] and graph processing [209]. To the best of our knowledge, there is no prior survey of dedicated hardware accelerators for NFs.…”

Section: E Dedicated Acceleratorsmentioning

confidence: 99%

Hardware-Accelerated Platforms and Infrastructures for Network Functions: A Survey of Enabling Technologies and Research Studies

2020

View full text Add to dashboard Cite

In order to facilitate flexible network service virtualization and migration, network functions (NFs) are increasingly executed by software modules as so-called "softwarized NFs" on General-Purpose Computing (GPC) platforms and infrastructures. GPC platforms are not specifically designed to efficiently execute NFs with their typically intense Input/Output (I/O) demands. Recently, numerous hardwarebased accelerations have been developed to augment GPC platforms and infrastructures, e.g., the central processing unit (CPU) and memory, to efficiently execute NFs. This article comprehensively surveys hardware-accelerated platforms and infrastructures for executing softwarized NFs. This survey covers both commercial products, which we consider to be enabling technologies, as well as relevant research studies. We have organized the survey into the main categories of enabling technologies and research studies on hardware accelerations for the CPU, the memory, and the interconnects (e.g., between CPU and memory), as well as custom and dedicated hardware accelerators (that are embedded on the platforms); furthermore, we survey hardware-accelerated infrastructures that connect GPC platforms to networks (e.g., smart network interface cards). We find that the CPU hardware accelerations have mainly focused on extended instruction sets and CPU clock adjustments, as well as cache coherency. Hardware accelerated interconnects have been developed for on-chip and chip-to-chip connections. Our comprehensive up-to-date survey identifies the main trade-offs and limitations of the existing hardware-accelerated platforms and infrastructures for NFs and outlines directions for future research.

show abstract

A Survey on Agent-based Simulation Using Hardware Accelerators

Cited by 38 publications

References 161 publications

Causality and Consistency of State Update Schemes in Synchronous Agent-based Simulations

Causality and Consistency of State Update Schemes in Synchronous Agent-based Simulations

Differentiable Agent-Based Simulation for Gradient-Guided Simulation-Based Optimization

Hardware-Accelerated Platforms and Infrastructures for Network Functions: A Survey of Enabling Technologies and Research Studies

Contact Info

Product

Resources

About