Improving the Performance of Pre-copy Virtual Machine Migration Technique

Abstract

Virtualization is the key technology running behind the cloud servers to deploy users services efficiently. So, when a user requests for cloud services, resources are allocated in the form of VM instance. With the tremendous growth in cloud services, the number of applications running over cloud servers is increasing. Therefore, the chances of server overloading and maintenance are quite often. This scenario is handled by live VM migration, that allows migration of VMs from one host to another with minimum disruption to cloud user services. Cloud hypervisors use pre-copy technique to perform the VM migration task. However, it suffers from page-level content redundancy problem, where complete VM memory page is transferred instead of modified contents of the page. Therefore, a technique is proposed which transfers only modified contents of VM memory page. The performance of proposed technique is evaluated with VMs running read-and write-intensive workloads configured on QEMU-KVM virtualized environment. The results obtained demonstrate that for read-intensive workload, compared to existing pre-copy VM migration scheme, our proposed technique minimize downtime by 69.4%, migration time by 73.5%, and reduces network traffic generated by 74.21%. Similarly, for write-intensive workload, downtime, migration time, and network traffic generated are reduced by 47.23, 49.16, and 51.37%. Thus, the proposed technique effectively improves the performance of VM migration scheme.