使用CUDA提升NCTUns網路模擬器的效能

Abstract

過去幾十年，許多研究學者提出關於利用多台電腦或多顆核心的分散式平行模擬的演算法，來提升模擬大型複雜網路環境的效能。近幾年，Graphics Processing Unit (GPU)因為其驚人的浮點運算能力以及其多重核心架構，提供了高效能及高密度的平行運算能力，並且能夠以更少的資源來處理更為複雜的計算問題。 在本篇論文中，我們利用GPU來提升NCTUns網路模擬器的效能。我們研究了NCTUns網路模擬器，發現無線網路實體層的相關運算適合交由GPUs來協助計算，並對NCTUns網路模擬器進行修改且量測效能提升的比例，最後提出本研究未來的發展性和需要改進的方向。

Since the past decades, researchers have proposed the parallel and distributed simulation method on multiple computers or multiple cores to increase the efficiency of simulating large-scale networks. In recently years, as a side product of the research of parallel and distributed computing, the Graphics Processing Unit (GPU) was introduced as a specialized microprocessor with amazing computing capability. With the GPUs’ highly parallel structure using the integrated multiple cores, GPUs promise higher performance and efficiency of a range of complex algorithms, as compared to those general-purpose CPUs. In this paper, we propose to use the GPUs in personal computers to improve the simulation efficiency of NCTUns network simulator. We discuss from different aspects of why and how the GPUs can be adopted by NCTUns network simulator for further performance enhancement. Also, we conduct a series of simulations to quantify the increment of the simulation efficiency through running the complex computations on the GPUs. Lastly, we conclude the paper with future development and direction for further enhancements.