Domain Project

We are delighted to announce that our paper, “Follow the Leader: Alternating CPU/GPU Computations in PDES,” has been accepted for presentation at the 38th ACM SIGSIM Conference on Principles of Advanced Discrete Simulation (SIGSIM PADS ’24) in Atlanta, GA.

The paper addresses the challenge of optimizing discrete-event simulations (DES) on heterogeneous computing platforms that include both CPUs and GPUs. Traditional static device choices between CPU and GPU often lead to suboptimal performance due to varying computational demands. Our research introduces a dynamic execution scheme that alternates between CPU and GPU, aiming to approximate an optimal device selection in real-time.

The key goal of the paper is to demonstrate a method for dynamically selecting the optimal computing device (CPU or GPU) based on ongoing performance measurements. This method includes a “follow-the-leader” approach, where a leading device periodically saves its state and is then challenged by the other device. Performance forecasts based on these challenges guide the choice of the next leader device, optimizing both execution time and energy consumption.

The results of our experiments show significant reductions in execution time without increasing energy consumption compared to static device choices. This approach has the potential to greatly enhance the efficiency of simulations in various fields, leveraging the strengths of both CPUs and GPUs dynamically.

We look forward to presenting our findings and discussing their implications with the community at SIGSIM PADS ’24.