Energy Consumption Based Multi-VM Scheduling System

The large-scale multicomputer system provides rapid processing power and strongstorage capacity. However, it also brings huge energy consumption and maintenance costs.With the help of computing system virtualization technology, multiple virtual machines areable to run concurrently on one single physical machine. The virtual machines experiencedynamic change in demand of physical resources, which incurs difference of load intensityon physical machines. Some physical machines keep in a relatively low load level. Theresources of computing system are not fully utilized, which provides conditions forconsolidation of virtual machines.The energy consumption based multi-VM scheduling system solves the aboveproblems in two ways: by adjusting the load intensity on physical machines andconsolidating the virtual machines on low-load physical machines. Following the steps ofenergy based multi-VM scheduling algorithm, the scheduling system monitors the resourceusage of virtual machines and uses the CPU utilization-based power model of physicalmachines to figure out the proper distribution of virtual machine to physical machine whichnot only meets the resource demands of virtual machines but also minimizes powerconsumption of the computing system. The overhead of distribution is paid close attentionto that the times of live migration involved is minimum.The experimental results obtained in10different system load levels shows that under3different virtual machine configurations, with the distribution of energy consumptionbased multi-VM scheduling algorithm the number of active work nodes decreases by31%,16%and18%respectively; the average power consumption of the system reduces24%,17%and9.7%respectively; the average memory utilization of the system increases to40.7%,65.1%and47.3%respectively; the times of live migration are reduced to theminimum number that needed to consolidate the idle work nodes.