Thread-Based Live Checkpointing of Virtual Machines

Virtual machine check pointing is the mechanism to save virtual machine state to a file for later recovery. Traditional check pointing mechanisms can suffer a long delay and cause a long disruption of services since they have to stop virtual machines to save state, which could be large. In this study, a novel Thread-based Live Check pointing (TLC) mechanism is proposed. This mechanism leverages the pre-copy live migration mechanism introducing a checkpoint thread, which is responsible for the majority of the check pointing activities. While the checkpoint thread is saving the virtual machine state to persistent storage, the virtual machine thread is allowed to progress with normal execution. However, the virtual machine thread will be periodically interrupted to incrementally copy dirty memory pages to a hash table. The interruptions will occur until the final stage of check pointing is reached. This approach is implemented in KVM and its performance evaluations are conducted using NAS parallel benchmarks. Experiments show that this approach can provide high levels of virtual machine responsiveness during check pointing. It can also reduce the check pointing overheads to as low as 0.53 times of that of the traditional approach, when operating on a virtual machine running memory intensive workloads.