Virtual Machine Migration as a Fault Tolerance Technique for Embedded Real-Time Systems

Virtualization's architectural abstraction and encapsulation of guest systems in virtual machines facilitate migration, but existing real-time virtualization solutions are characterized by a static mapping of virtual machines to processors. This work studies migration of virtual machines with real-time constraints on homogeneous multiprocessor architectures as a service restoration in response to hardware faults. The migration policy respects real-time requirements and minimizes and predicts deadline misses based on a preceding comparison of downtime caused by the migration and slack-based computation of the virtual machine's maximum affordable downtime. The distributed design is characterized by a communication between the paravirtualized operating system and the hypervisor in order to provide the required scheduling information. The overhead regarding memory footprint, execution times, and paravirtualization effort is analyzed. The evaluation identifies ranges for virtual machine size and timing characteristics for which the approach is feasible. A reliability analysis based on a combinatorial model is used to quantify the impact of migration on reliability.