CNS Core: Small: Toward Real-Time Stream Processing in Edge Devices

及时分析实时传感器数据流对于智能健康、交通和能源等物联网（IoT）的关键应用至关重要。尽管先进的流处理引擎（SPE），如Apache Storm、Flink和Spark Streaming，在云中提供了强大的流处理框架，但是将传感器数据发送到SPE以在广域网上进行分析可能会导致许多最后期限的错过，并在核心Internet中造成瓶颈。一个可行的替代方案是在边缘设备中进行实时传感器数据流处理；然而，在这些设备中使用有限的可用资源来支持时间约束是具有挑战性的。实时调度理论并不直接适用，因为它与数据语义无关，并且通常基于最坏情况的可预测性假设，这在边缘设备中过于悲观且资源效率低下。随着物联网设备数量和数据量的快速增长，这个问题变得越来越严重。提出的工作旨在通过研究边缘软实时流处理的经济有效方法来弥合日益扩大的差距。该项目探索了调度、传感器流处理和负载共享的新方法，以显著减少截止日期错过、通信和计算资源消耗，同时提高实时流处理的可靠性。与最先进的spe相比，该研究有望通过大幅提高实时流处理的及时性和可靠性，以更少的资源消耗，为产生巨大社会影响的重要物联网应用（如医疗保健、交通运输和能源领域）提供一种使能技术。研究者将使用选定的研究结果继续教育和推广工作，包括广泛传播该项目将产生的出版物和代码，开发实时流处理的新课程和教材，招募代表性不足的学生群体参与该项目，并鼓励年轻一代学习计算机科学并在工业界和学术界从事职业。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Timely analysis of real-time sensor data streams is essential to key applications in the Internet of Things (IoT), such as smart health, transportation, and energy. Although advanced stream processing engines (SPEs), such as Apache Storm, Flink, and Spark Streaming, provide powerful stream processing frameworks in a cloud, sending sensor data to the SPE for analysis over the wide area network may incur many deadline misses and create bottlenecks in the core Internet. A viable alternative is real-time sensor data stream processing in edge devices; however, it is challenging to support timing constraints using limited resources available in such devices. Real-time scheduling theory is not directly applicable, since it is agnostic to data semantics and usually based on worst-case assumptions for predictability that would be too pessimistic and resource inefficient in edge devices. The problem is becoming increasingly serious as the number of IoT devices and data volume increases rapidly. The proposed work aims to bridge the widening gap by investigating cost-efficient approaches for soft real-time stream processing at the edge. This project explores novel approaches to scheduling, sensor stream processing, and load sharing to significantly decrease deadline misses and communicational as well as computational resource consumptions, while enhancing the reliability of real-time stream processing. The research is expected to provide an enabling technology for important IoT applications with great societal impacts, such as those in healthcare, transportation, and energy that produce immense real-time sensor data streams, by substantially improving the timeliness and reliability of real-time stream processing with less resource consumptions compared to state-of-the-art SPEs. The investigator will use select research results to continue education and outreach efforts that include broadly disseminating publications and code that will be produced by this project, developing new courses and teaching materials on real-time stream processing, recruiting underrepresented groups of students to work on the project, and encouraging the younger generation to study computer science and pursue careers in industry and academia.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.