高效能可扩展边缘计算体系结构关键技术研究

万物互联和人工智能对大数据实时处理的计算需求，对已有计算模式和系统架构带来巨大挑战，边缘计算正在成为新兴万物互联应用的支撑架构。本课题针对边缘计算环境中，系统结构可扩展性不强、算力和计算效能不足、存储和I/O性能不高、应用需求多样等挑战，从三方面开展创新性研究：1)在动态可扩展系统架构设计方面，研究基于软件定义的异构计算资源池化管理技术、边缘应用服务能耗建模与估测方法；2)在边缘节点的计算效能增强方面，主要研究满足边缘节点功耗和算力约束的人工智能FPGA加速器体系结构；3)在边缘存储和I/O性能优化方面，研究边缘数据的高效预取与自适应缓存替换策略、面向云-边-端协作缓存的联合学习策略。研究成果将为研制高效能可扩展边缘计算系统提供基础构建方法和技术支撑。

With the rapid development of the Internet of Everything and artificial intelligence, the ever-expanding data scale and the real-time data processing have brought huge challenges to the existing computing models and systems. Edge computing is becoming the supporting infrastructure for the emerging Internet of Everything applications. Aiming at the challenges of weak structure scalability, insufficient computing power and efficiency, limited storage and I/O performance, and diverse application requirements in the edge computing environment, this project intends to carry out innovative research from three aspects: (1) In the aspect of dynamic and scalable system architecture design, we mainly study software-defined heterogeneous computing resource pooling management technology, and edge energy consumption modeling and assessment strategy. 2) In terms of enhancing the computing efficiency of edge nodes, we focus on the FPGA accelerator architecture for edge AI that meets the constraints of edge node power consumption and computing power. 3) For storage and I/O performance optimization, we study structure optimization of efficient prefetching and adaptive cache replacement strategies, and a cooperative learning strategy for cloud-edge-end collaborative cache. The prototype system will be designed and implemented for verification. The research results will provide basic construction methods and technical support for the development of high-efficient and scalable edge computing systems.