Predicting WCET of automotive software running on virtual machine monitors

Virtualization is attracting significant interest in the automotive industry because it enables a highly secure and reliable computing environment. More importantly, virtualization maintains the same operating environment for legacy automotive software while exploiting the benefits of widely adopted multicore platforms. To exploit the virtualization technology in an automotive system, it is important to predict the WCET of an automotive application running on a virtual machine monitor (VMM). Unfortunately, the task is challenging because of difficulties in analyzing complicated interactions between a VMM and a guest OS. There are no known attempts to predict the WCET of an application in such an environment. In this paper, we propose a hierarchical and parametric WCET prediction framework. We divide the problem into two subproblems. First, we model the WCET of an application as a function of WCETs of system calls provided by a guest OS. Second, we model WCETs of a system call as a function of WCETs of VMM services. To establish this framework, we clearly identify the places and times of VMM services invoked during the execution of an application. At the time of deployment, the WCET of an application is instantiated by composing the WCET models altogether. We have performed experiments with the proposed framework by predicting the WCETs of sample programs on various virtual and real machine platforms. These experimental results effectively demonstrate the viability of the proposed framework.