Andrea Claudi, ADB Broadband
Agostina Fioravanti, Università Politecnica delle Marche
Luca Spalazzi, Università Politecnica delle Marche
Gilberto Taccari, Università Politecnica delle Marche
Leonardo Taccari, Università Politecnica delle Marche

Industrial real-time systems are widely used in various application domains. In these fields, flexibility and cost are the two most important metrics to measure, from the non-functional point of view, the quality of a real-time system. To meet the expectations of the industry, research on real-time systems is focusing on two main directions: virtualization and multicore real-time scheduling.

Virtualization technologies allow to deploy multiple real-time systems on common computing platforms, instead to use a single host for each of them. This is a huge benefit in terms of flexibility: for example, it allows to stop a platform (and thus save power) when it is not needed, moving real-time virtual machines (VMs) to other platforms. Integrating virtualization technologies with multicore platforms while meeting the real-time requirements is not easy, and poses significant research challenges.

The use of virtualization on embedded systems dates back to recent years. Indeed, embedded systems are getting more and more powerful and recently ARM processors implemented hardware-supported virtualization. Various open-source and high-performance hypervisors, KVM and Xen in particular, are being ported to the ARM processors with virtualization extensions and official support is also present in mainline Linux kernel. After a brief look to the ARM hardware-supported virtualization extensions and the current status of Xen and KVM hypervisors for ARM CPUs and real-time solution for them, performances of real-time processes under VM host and guests will be analyzed via ad-hoc benchmarking tools and they will be compared to the native performance (eg. without running an hypervisor).