On-Detector Processing and Data Streaming
On-Detector Processing and Data Streaming
复制标题
探测器上处理和数据流
DOI：
10.1080/08940886.2015.1013416
复制
发表时间：
2015
期刊：
影响因子：
--
通讯作者：
I. Johnson
中科院分区：
文献类型：
--
作者：
I. Johnson
x射线探测器技术的进步使同步加速器源和自由电子激光器的许多学科取得了新的科学突破。检测器系统的增强功能通过更高的帧速率扩展动态范围，增加像素计数，增加整体帧速率也提高了数据速率。如今，x射线应用使用最先进的百万像素探测器，帧率为100赫兹到10千赫，产生的原始数据速率可以超过100 Gbps。考虑到这些大的数据速率，简单地将原始数据写入本地磁盘并从磁盘读取以供后续分析变得不那么有吸引力，而且通常是不切实际的。现代架构开始通过检测器信号处理、早期无损压缩和直接数据流到下游CPU或GPU集群来解决数据源上的数据洪流，其中内联分析可以为科学家提供实时实验反馈。
Advancements in X-ray detector technology have enabled new scientific breakthroughs in many disciplines at synchrotron sources and free-electron lasers. Enhancements to detector systems that extend dynamic range through higher frame rate, increase pixel count, and increase overall frame rates also boost data rates. Today, X-ray applications utilize state-of-the-art mega-pixel detectors at 100s Hz to 10s of kHz frame rates, creating raw data rates that can exceed 100 Gbps. Simply writing raw data to local disks and reading it back from the disk for subsequent analysis is becoming less attractive and often impractical, given these large data rates. Modern architectures are beginning to tackle the data deluge at the source with on-detector signal processing, early lossless compression, and direct data streaming to downstream CPU or GPU clusters, where inline analysis can provide scientists with real-time experimental feedback.