基于跨层机制的嵌入式虚拟化技术优化方法研究

Embedded virtualization is an emerging technology in the application field, which can reduce the develop cost, increase security and hardware utilization. However, the general-purpose virtualization frameworks are not designed for the resource constraints environment of embedded system. In this way, it is difficult to fulfill the strict constraint on real-time response, energy consumption and performance isolation if the virtualization framework is directly ported to the embedded hardware. Besides, because of the complexity of virtualization software stack and the independent design of different layers, it is impossible to solve the problem by a single layer optimization. In this project, a cross-layer approach is proposed to break the isolation between the original software layers. The cross-layer information channels and control interfaces is introduced to optimize the scheduler design. An efficient scheduling strategy is designed based on the context information and new control interface, so that the collaboration between multiple layers is fully considered. Specifically, the project will focus on the optimization of embedded virtualization in the following aspects. (1)Energy consumption optimization of virtual machine scheduling on the heterogeneous multi-core embedded processor based on the response time information from the application layer. (2)Real time optimization of I/O request scheduling based on the cross-layer information of the I/O software stack and the control interfaces. (3)Resource allocation isolation optimization based on device control strategy implemented by programmable hardware. According to the research of this project, an efficient optimization scheme can be provided to improve the performance and usability of embedded virtualization. The research result of this project can be used to improve the industry application of the embedded virtualization and the optimization methods are also valuable in other application area of virtualization technology.

嵌入式虚拟化技术能够降低开发成本、提升硬件利用率和安全性，具有广阔的应用前景，但由于通用虚拟化框架缺少针对嵌入式系统资源受限环境的设计，导致在实时性、能效、性能隔离等方面存在不足。而且虚拟化软件层次结构复杂，各层设计相对独立，使得单一层次的优化难以有效解决问题。本项目旨在利用跨层机制取得突破，打破原有的软件层次隔离，为调度决策提供更完整的上下文信息和更精确的控制接口，从而实现多层协同的高效调度机制，在不同层面对嵌入式虚拟化技术进行优化。重点研究以下内容：利用应用层的响应时间信息优化虚拟机调度层的嵌入式异构多核处理器调度以提高能效；利用I/O软件栈各层信息和控制接口进行请求的调度优化以提高实时性；利用可编程硬件在设备控制层实现调度层的资源分配策略以提高隔离性。本项目的研究成果能够提升嵌入式虚拟化技术的性能和可用性，推动其产业应用，并能促进虚拟化技术在其他领域的应用优化。

虚拟化技术能够降低硬件的复杂度、提高软件的复用率，有效提升系统的开发效率和资源利用率。但是当虚拟化技术进入嵌入式领域时遇到了新的挑战。虚拟化技术通过层次式的抽象设计屏蔽底层细节以实现软件的复用，而嵌入式系统却需要依赖对硬件的细节信息和直接控制以保证系统的高能效、低延时等特性。为解决这一问题，本项目引入跨层机制，打通原有软件的层次化设计，弥补虚拟机带来的语义鸿沟，使得虚拟机处理器、IO请求在调度的过程中能够更具针对性，从而提高系统的能效和实时性。具体来说，(1)项目实现了虚拟机内部应用响应时间的延迟信息用于指导在异构多核实理器上的虚拟机处理器的调度，从而为虚拟机选择适合其算力需求的虚拟处理器；（2）项目实现了IO软件路径上从系统调用到物理硬件的全程元数据贯通传输，打通了虚拟机操作系统、虚拟设备、宿主机操作系统的信息通路，使得IO请求在硬件层调度时更具针对性，在高负载环境下能够将高优先级线程的最差请求确认时间从240毫秒提升到2毫秒；（3）项目实现了硬件状态的及时通知，可以使得虚拟机能够在数据传输的同时可以对设备状态做出响应，同时使得宿主机可以对虚拟机进行快速准确的细粒度控制，在实际使用中实现了对IO请求的及时阻断以避免在紧急状态下的数据洪泛。在上述研究的基础上，本项目探索了异构指令集处理器的虚拟化技术，利用动态二进制翻译实现统一的指令集接口，并使用分布式计算的技术手段将分立的异构处理器组成一个协同工作的集群，通过数据分块、数据预取、线程调度等方式优化性能，从而为单个进程提供了一个接口统一的分布式异构计算平台，在理想状态下获得近乎线性的加速比。