A Research On Real-time Responsiveness In Embedded Virtualization System

In Recent years embedded world has been undergoing a shift from traditional single-core processors to processors with multiple cores. However, this shift poses a challenge of adapt-ing legacy uniprocessor-oriented real-time operating system (RTOS) to exploit the capability of multi-core processor. In addition, some embedded systems are inevitably going towards the direction of integrating RTOS with general-purpose operating system (GPOS) , as the combi-nation of the two has the potential to provide not only timely and deterministic response but also a large application base. Virtualization technology, which enables several virtual ma-chines to run on single hardware platform, is therefore a promising solution to above men-tioned issues. In this regard, armed with virtualization support, RTOS and GPOS are able to run together on top of separate virtual machines. This setup can utilize multi-core more effi-ciently, for instance, different virtual machines can be designated to different physical cores.In embedded systems, the real-time performance, especially responsiveness to external events, is often a key factor to their wide applications. With the introduction of virtualization to the embedded systems, however, the responsiveness of the RTOS running on top of a vir-tual machine is prone to be adversely affected by the virtualization software layer. Thus this fact poses two questions: First, how much does the virtualization software layer affect the re-sponsiveness. Second, what level of responsiveness can be achieved by the embedded virtual-ization systems.To answer above mentioned questions, in this thesis we propose an embedded real-time virtualization architecture based on Kernel-Based Virtual Machine (KVM), in which VxWorks and Linux are combined together as guest RTOS and guest GPOS respectively. Fir- st, we analyze how KVM influences the timer interrupt response times of the guest RTOS by studying the mechanisms of both interrupt virtualization and time virtualization of KVM. We then analyze several real-time performance tuning methods on the host Linux, aiming at eas-ing the extra overheads incurred by KVM. Finally, we conduct several experiments to evalu-ate the responsiveness of the guest VxWorks. The experimental results show that after apply-ing performance tuning methods, sub-millisecond interrupt response latency can be achieved on the guest VxWorks.