A GPU-Based Simulation Kernel within Heterogeneous Collaborative Computation on Large-Scale Artificial Society

Abstract

Abstract-The graphic processing unit (GPU) gets strong computing ability with relatively low energy and money consumption, it has been widely used in the field of large-scale simulation and computation. Among which the CPU-GPU heterogeneous collaborative computing model has become an effective ways to solve the simulation performance of large-scale artificial society. But there are lots of problems in GPU-based ABS. The paper proposes a GPU-based conservative parallel discrete event simulation algorithm for ABS. We reorganize data structure for CPU/GPU-based heterogeneous collaborative parallel simulation, design a GPU kernel scheduling algorithm based on conservative time synchronization strategy, propose an efficient organization and scheduling algorithm for simulation event and improve execution efficiency of conservative time synchronization algorithm through the optimization of large-scale parallel time reduction algorithm. Finally, we analyze the algorithms proposed and the GPU-based simulation kernel with GameOfLife model, up to 11.2x speedup is obtained compared to CPU. Index Terms-Artificialsociety, ABS, CPU/GPU, heterogeneous collaborative simulation, conservative parallel.

A Survey on Agent-based Simulation Using Hardware Accelerators

“…A number of works avoid the need for a global PQ holding all future events. Instead, the set of events is considered jointly in an unsorted fashion [153], split by model segment [142] or simulated entity [179,5,98], split according to a fixed policy [124,166], or split randomly [108]. To determine the events that can be executed without violating the simulation correctness, a parallel reduction is performed to determine the minimum timestamp among the events.…”

“…As building blocks for time advancement in a discrete-event fashion, parallel reduction and bitonic sorting are commonly used in GPU-and FPGA-based simulation [123,163,179,142,85]. We discuss these two operations jointly due to their structural similarities.…”

“…Although the simulation results may be affected by their approach, the authors show that for a queueing network simulation, error bounds can be given. Other works assume a minimum time delta between an event and its creation (lookahead ) to guarantee the correctness of the simulation results [140,5,179]. If lookahead is available, a window can be determined within which events are independent, allowing for parallel execution without affecting the simulation correctness.…”

Performance Evaluation of Priority Queues for Fine-Grained Parallel Tasks on GPUs

“…In their approach, each new event is placed in a randomly selected column, linked lists connecting each LP's events in timestamp order. Both Zhen et al [34] and Andelfinger et al [35] store each LP's events in a separate array. Parallel access to an LP's queue is performed using atomic operations.…”