A Two-Level Virtual Machine Self-Reconfiguration Mechanism for the Cloud Computing Platforms

Cloud computing is a new model and technology that leverage the efficient pooling of on-demand, self-managed virtual infrastructure. Virtualization packages the applications in the form of Virtual Machine (VM) and provides significant benefits by reconfiguring the VMs dynamically. VM reconfiguration is hard and complicated, and existing work addressed the problem with diverse objectives by answering the questions of when to reconfigure, which VMs should be reconfigured and where to host the VMs. However, we found that the runtime reconfiguration affects the total costs significantly. Then we propose a two-level runtime reconfiguration mechanism to automate the operations with the objective of minimizing the costs. The mechanism includes the local adjustment and the parallel migration. Employing the local adjustment, VMs on a same server can be reconfigured in a time-division multiplexed way based on the load trend prediction, which can avoid the unnecessary VM migrations. Nevertheless, VM migration is inevitable when the server is overloaded. Considering the conflict between reducing the migration cost and minimizing the performance interference, we propose a VM parallel migration strategy and map it to the max matching problem of the bipartite graph. We implement a framework based on Xen and evaluate the mechanism with a preliminary experiment. The results show that this two-level self-reconfiguration mechanism is effective in reducing the VM runtime reconfiguration costs.