Live Migration Analysis for Multitasking Virtual Machines under Memory Stresses

Virtual machine live migration is an important. mechanism that enables virtual machines to move from source hosts to their target hosts in seamless way. By the aid of live migration, energy saving and load balancing on cloud computing platforms can be easily achieved. However, a long service downtime at the stop -and -copy stage of live migration may violate the service level agreement cosigned by the providers and customers. The downtime can be reduced if the memory space available to tasks is shrunk before a migrating virtual machine enters the stop -and -copy stage. Those pages which are swapped out due to the shortage of available memory will be swapped in after the virtual machine has successfully resumed at the target host. Page -swapping degenerates the performance of a virtual machine because of resource underutilization. Multitasking can be used to compensate for the performance loss caused by intensive page swapping. In this paper, we analyse the performance of a multitasking virtual machine which migrates under memory stress. The perliwmance metric is the time taken by a multitasking virtual machine to complete all the concurrent tasks which start at a host and finish at another. Our experimental results show that the performance of a multitasking virtual machine with live migration demand can be improved if tasks arc executed concurrently under appropriate memory stress settings.