Acceleration of atmospheric Cherenkov telescope signal processing to real-time speed with the Auto-Pipe design system
Acceleration of atmospheric Cherenkov telescope signal processing to real-time speed with the Auto-Pipe design system
复制标题
使用 Auto-Pipe 设计系统将大气切伦科夫望远镜信号处理加速至实时速度
DOI：
10.1016/j.nima.2008.06.047
复制
发表时间：
2008
影响因子：
1.4
通讯作者：
R. Chamberlain
中科院分区：
文献类型：
--
作者：
Eric J. Tyson;J. Buckley;M. Franklin;R. Chamberlain
用于高能γ射线天文的大气切伦科夫成像技术是研究高能宇宙的一项重要新技术。目前的实验数据速率为120 TB/年，占空比约为10%。在未来，更敏感的实验可能会产生高达1000 TB/年。这些实验的数据分析任务需要接近实时地保持这种数据速率。这种数据分析是具有非常高性能要求的流应用的典型示例。这类应用通常受益于使用非传统的计算方法，包括使用专用硬件（FPGA和ASIC）或复杂的并行处理技术。然而，设计、调试和部署到这些架构是困难的，因此它们没有被天体物理学社区广泛使用。本文介绍了自动管道设计工具集，已开发，以解决许多困难，利用复杂的流式计算机架构，这样的应用程序。Auto-Pipe集成了高级协调语言、功能和性能仿真工具，以及将应用程序部署到复杂架构的能力。使用Auto-Pipe工具集，我们实现了成像切伦科夫数据分析应用程序的前端部分，适用于实时或离线分析。该应用程序基于VERITAS实验的数据运行，并展示了Auto-Pipe如何极大地简化各种平台的性能优化和应用程序部署。我们证明了一个传统的软件方法的32倍使用FPGA解决方案和3.6倍使用基于多处理器的解决方案的性能提高。
The imaging atmospheric Cherenkov technique for high-energy gamma-ray astronomy is emerging as an important new technique for studying the high energy universe. Current experiments have data rates of ≈20TB/year and duty cycles of about 10%. In the future, more sensitive experiments may produce up to 1000TB/year. The data analysis task for these experiments requires keeping up with this data rate in close to real-time. Such data analysis is a classic example of a streaming application with very high performance requirements. This class of application often benefits greatly from the use of non-traditional approaches for computation including using special purpose hardware (FPGAs and ASICs), or sophisticated parallel processing techniques. However, designing, debugging, and deploying to these architectures is difficult and thus they are not widely used by the astrophysics community. This paper presents the Auto-Pipe design toolset that has been developed to address many of the difficulties in taking advantage of complex streaming computer architectures for such applications. Auto-Pipe incorporates a high-level coordination language, functional and performance simulation tools, and the ability to deploy applications to sophisticated architectures. Using the Auto-Pipe toolset, we have implemented the front-end portion of an imaging Cherenkov data analysis application, suitable for real-time or offline analysis. The application operates on data from the VERITAS experiment, and shows how Auto-Pipe can greatly ease performance optimization and application deployment of a wide variety of platforms. We demonstrate a performance improvement over a traditional software approach of 32x using an FPGA solution and 3.6x using a multiprocessor based solution.