Real-time performance evaluation of Linux ARM virtualization

In order to meet the time constraints of real-time tasks, embedded real-time systems are used to assign different real-time tasks to multiple processors. However, using multiple processors makes the system to consume more energy, harder to maintain and to grow in size and weight. In recent years, ARM CPUs have become increasingly common in embedded systems. With the improvement of its performance, more embedded systems tend to use only one high-performance processor to run all tasks. There is a growing demand for a solution to guarantee both isolation and time constraints of the tasks. We present our real-time virtual machine solution, which leverages the newest real-time scheduler of Linux and the hardware virtualization support of ARM CPUs. We take the maximum interrupt frequency of the virtual machine as one real-time performance metric. In addition, we consider the maximum execution frequency of a specific real-time task as the other metric. The results of our experiments demonstrate that the real-time performance of our virtual machine is roughly 23% of the native machine and the real-time performance of the virtual machine is linear with its CPU resource allocation ratio.