Accelerating sequential Monte Carlo method for real-time air traffic management
Accelerating sequential Monte Carlo method for real-time air traffic management
复制标题
用于实时空中交通管理的加速顺序蒙特卡罗方法
DOI：
10.1145/2641361.2641367
复制
发表时间：
2013
期刊：
影响因子：
--
通讯作者：
J. Maciejowski
中科院分区：
文献类型：
--
作者：
T. Chau;James Stanley Targett;Marlon Wijeyasinghe;W. Luk;P. Cheung;Benjamin Cope;A. Eele;J. Maciejowski
本文介绍了如何使用现场可编程门阵列（FPGA）来加速空中交通管理的序贯蒙特卡罗方法。一种新的数据结构引入的粒子流，使有效的评价的约束和权重。这种流数据结构的并行实现的设计，并提供了一个分析模型来估计我们的实现的性能和资源使用情况。我们比较我们的设计，CPU和GPU上的实现。我们展示了与具有8个线程的Intel Core i7-950 CPU相比的9.3倍速度和89倍能效提高，并展示了与具有448个内核的NVIDIA Tesla C2070 GPU相比的1.3倍速度和13.5倍能效提高。我们还估计了FPGA在未来场景中的性能，并表明FPGA能够实时控制的飞行器分别是CPU和GPU的15倍和2.8倍。
This paper presents how field-programmable gate arrays (FPGAs) are used to accelerate the Sequential Monte Carlo method for air traffic management. A novel data structure is introduced for a particle stream that enables efficient evaluation of constraints and weights. A parallel implementation for this streaming data structure is designed, and an analytical model is provided for estimating the performance and resource usage of our implementation. We compare our design to implementations on CPU and GPU. We show 9.3 times speed up and 89 times improvement in energy efficiency over an Intel Core i7-950 CPU with 8 threads and demonstrate 1.3 times speed up and 13.5 times improvement in energy efficiency over an NVIDIA Tesla C2070 GPU with 448 cores. We also estimate the performance of FPGA in future scenario and show that FPGA is able to control 15 times and 2.8 times more aircraft than CPU and GPU in real-time respectively.