2. Contribution

This paper presents how we can run a real-time guest OS over a Xen hypervisor.

At first, we need to deal with the scheduling problem.

All the scheduling inside a guest OS should be preserved in terms of real-time schedulability.

Namely, schedulability analysis methodology has to be provided so that we can analytically determine the system is able to finish all the real-time tasks within a given deadlines.

Secondly, the hypervisor should be able to deliver physical interrupt event to guest in a timely manner.

It is essential to provide predictable I/O processing, and time-bounded interrupt handling.

We incorporate the real-time scheduling theory into a real system, which provides the fundamentals of real-time scheduling.

Compositional scheduling theory can be applied so that the virtual machine scheduling at hypervisor guarantees the real-time behavior.

Unfortunately, the theory is not directly applicable to the virtual machine scheduling because it induces a significant overhead.

We provide an algorithm that finds the optimal interface while preserving the real-time properties.

In addition, we elaborate the interrupt processing so that the physical interrupt can be immediately delivered to the guest OS side.

After the event delivery, the real-time guest OS can process it with the minimal performance overhead.