面向众核的空分系统结构研究

Exploring Non-Uniform Memory Access Architecture for Multi-core Platform

虚拟机技术通过操作系统级的并行能够一定程度上适应硬件的并行发展。但是其层叠性能损失和能耗增加使得其应用有限。本文中分析了当前虚拟机技术存在的缺陷,引入了空分复用的思想,并重新设计了一个面向众核平台的系统结构,通过将不同功能部分绑定到特定处理器上,通过处理器间的中断来完成功能调用。这种机制避免了时分复用技术带来的切换开销,通过处理期间显式消息传递代替了难以确定执行时间的cache一致性协议,并针对NUMA结构下,处理器地位不一致的情况,提出了相应的分配和迁移策略。

A virtual machine can adapt to the parallel development oI hardware to a certain extent mrougn the oper ating systemlevel parallelism. However, it causes the loss of stack performance and increases energy consumption. This paper analyzes the shortcomings of the current virtual machine, introduces the space division multiplexing, and redesigns the architecture of a multicore platform by binding different functions to specific parts of the processor, and the break between processors completes function calls. This mechanism avoids switching costs caused by the time division multiplexing. The explicit message passing during processing takes the place of cache coherence proto cols because it is difficult to determine the execution time. Because the processor status is inconsistent with the NU MA architecture, we propose the corresponding distribution and transfer strategies.