CSR-DMSS, SM: Operating System Abstractions for Modern Hardware

点击翻译按钮获取中文摘要

As CPUs move towards increasing parallelism and new memory technologies are developed, operating system abstractions of hardware have changed little in the past 30 years. However, computer architecture is moving towards heterogeneous multi-core architectures that dynamically vary the number, performance, power, and reliability of processing elements. Memory technologies such as flash, magnetoresistive and phase-change RAM promise large quantities of fast, non-volatile memory. This project seeks to identify current OS abstractions that will be outdated on future processing platforms, invent new abstractions that more accurately capture the range of platform behavior, and implement these new abstractions in the context of existing operating systems. The research focuses on extensions: (1) abstracting hardware support for synchronization, such as transactional memory, thread queues, or full/empty bits, (2) abstracting heterogeneous multi-core processors, such as processors that have differing instruction sets or that can be fused together to form more powerful processing elements, (3) exposing non-volatile memory to programmers, and (4) abstracting microarchitectural contention for memory bandwidth, cache space, and shared processing elements to reduce the complexity of scheduling algorithms. The research will lead to higher-performing systems, due to both faster persistent storage and better use of execution resources, more reliable systems, due to faster persisting of data, and more programmable systems, due to general abstractions of specific hardware. In addition, the OS and architecture research communities will benefit from better software support for new hardware technologies. This work is inherently interdisciplinary and students will benefit from exposure to new research in other fields.