An Efficient GPU Implementation for Large Scale Individual-Based Simulation of Collective Behavior

Abstract

In this work we describe a GPU implementation for an individual-based model for fish schooling. In this model each fish aligns its position and orientation with an appropriate average of its neighbors positions and orientations. This carries a very high computational cost in the so-called nearest neighbors search. By leveraging the GPU processing power and the new programming model called CUDA we implement an efficient framework which permits to simulate the collective motion of high-density individual groups. In particular we present as a case study a simulation of motion of millions of fishes. We describe our implementation and present extensive experiments which demonstrate the effectiveness of our GPU implementation.

Overview of Case Studies on Adapting MABS Models to GPU Programming

“…for the study of flocking [8], crowd [24], traffic simulations [27] or autonomous navigation and path planning algorithms [4]). …”

Scaling-up spatially-explicit ecological models using graphics processors

“…The requirement for implementation of these models on the GPU is that on each individual or metacommunity, the same calculation is performed in parallel (e.g., not sequential). Individual-based models of schools of fish or migratory organisms have successfully been implemented on GPUs, as a similar movement algorithm is applied to all individuals (Erra et al, 2009;Guttal and Couzin, 2010). It is most efficient from computational viewpoint if individuals that interact with each other in the simulated group (e.g., are in each other's neighborhood) are also in proximity in the computer's memory, the individuals need to be continuously reordered in the computer's memory to match changes in the organisms coordinates (Erra et al, 2009).…”

“…Individual-based models of schools of fish or migratory organisms have successfully been implemented on GPUs, as a similar movement algorithm is applied to all individuals (Erra et al, 2009;Guttal and Couzin, 2010). It is most efficient from computational viewpoint if individuals that interact with each other in the simulated group (e.g., are in each other's neighborhood) are also in proximity in the computer's memory, the individuals need to be continuously reordered in the computer's memory to match changes in the organisms coordinates (Erra et al, 2009). This required a complex reordering algorithm, and hence implementation of these IBM models on GPUs is far more complex that the grid models that we discussed here.…”

Behavioral Spherical Harmonics for Long-Range Agents’ Interaction

“…The cell position is then hashed to give it a unique, sortable, consecutively-numbered value and stored a key-value pair with the agent's id as key and the cell position hash as value. This approach has been largely used in literature [5,6,18,8]. Our implementation of the spatial grid is similar to the NVIDIA particle simulation implementation [10]: steps 1-4 and 6 are essentially the same.…”

“…Researchers and practitioners of agent-based simulations and modeling have for a long time investigated the use of parallel implementations targeting a wide range of architectures, including multi-cores [15,17], GPUs [6,19], and distributed memory architectures [4,2,3]. Collision avoidance algorithms have been investigated by several ABS systems, where the motion of each agent is typically governed by some high-level formulation and local interaction rules (e.g., collision avoidance).…”