When a backup VM is not occupied by its dedicated service VM, it can be used by a failed service VM, which is called time sharing of backup VMs.
The time sharing of backup VMs can be achieved by starting a specific application on demand.
In this section, we first present our simple and effective heuristic time sharing policy.
If each backup VM is regarded as a bandit, the time sharing of backup VMs turns out to be a RMAB process.
With the known failure and restoration models of primary VMs, the objective of the heuristic time sharing policy is given by:
Before starting the heuristic time sharing policy, the state-transition matrices of the backup VMs and the initial belief state vector [OMEGA](1) are determined according to the failure and restoration models of the primary VMs.
For performance evaluation of the heuristic time sharing policy, we propose four metrics: backup assurance rate, idle time utilization of backup VMs, utilization of backup VMs and policy regret.
