CSR: Small: Collaborative Research: Efficient Exploitation of Heterogeneous Memory through OS/Compiler Support

内存是计算机的关键部件。现代并行体系结构开始使用异质存储器来改善存储器子系统的延迟和带宽。通常，异类存储系统由快组件和慢组件组成，并且需要明确的软件管理。它给程序员和编译器带来了独特的负担，特别是在高性能计算(HPC)和大数据分析领域。这些挑战对计算机系统的发展至关重要，因此也是未来科学发现、经济繁荣和国家安全的关键。为了解决利用异质内存的这些挑战，本项目的目标是联合设计编译器和操作系统(OS)：它探索了i)支持异构性感知数据管理的新编译器设计和ii)用于高效数据移动和公平内存共享的新OS设施。总体而言，该项目是迈向长期愿景的垫脚石--通过紧密集成编程和操作系统支持来驯服新兴的异类硬件。

Memory is a key component of computers. Modern parallel architectures begin to employ heterogeneous memory to improve both latency and bandwidth of the memory subsystem. Typically, a heterogeneous memory system consists of a fast and a slow component and requires explicit software management. It puts unique burdens on programmers and compilers, especially in the domains of high-performance computing (HPC) and big data analytics. These challenges are crucial to the evolution of computer systems and therefore are key to future scientific discovery, economic prosperity, and national security.To address these challenges in exploiting heterogeneous memory, this project targets co-designing the compiler and the operating system (OS): it explores i) a new compiler design that supports heterogeneity-aware data management and ii) new OS facilities for efficient data move and fair memory sharing. Overall, this project serves as a stepping stone towards a long-term vision -- taming emerging heterogeneous hardware by tightly integrating programming and OS support.