Research On Key Issues Of The Virtual Machine Consolidation

To improve resources usage and energy efficiency of server physical in datacenters, we can through consolidate virtual machines (VMs) to implement. MigratingVMs in real time between physical machines (PMs) and dynamically switch idlephysical machines to low power mode are important ways of dynamic consolidation.In this paper, we focus on the data transmittal pattern, the forecast of resourceusage and the relocation of VMs when exists VMs in state of overload or under load.First, we analyze the data transmittal pattern of typical monitor system and thecharacters of monitor information, and then propose an optimized transmittal patternof monitor data that manages the monitor information in an efficient way. Next, wepropose the Self-feedback Grey Markov model (SfGM) through analyzing charactersof Grey Model and Markoy theory, the N successive intervals of the VMs’ historicalmemory utilization as input and analyze the interrelation of these data to predict thefuture resources demand of PMs. In the end, monitoring resource occupancy andoperational state of VMs, PMs in datacenter, when the pre-establish conditions ofVMs of PMs are satisfied, we use the SfGM to predict the utilization of resource anddetermine whether execute VMs consolidation at present time point. If VMsconsolidation is needed, we will use the Heaviest PM the most suitable VMs firstRelocation (HPSVFR) Algorithm to relocate VMs in data centers.Simulation experiments with real-workload indicate that optimized transmittalpattern availably reduces the transmittal data amount and save the bandwidth ofnetwork. SfGM has the most precise ability to forecast that the mathematicalexpectation of SfGM is46.8%of GMM and only28.6%of GM (1,1). Comparedwith First Fit algorithm and Best Fit algorithm, HPSVFR takes least cost of all thatonly70%of other algorithms. The VMs consolidation architecture can effectivelydistinguish the utilization of resources in present datacenter. If there are some PMsand VMs stay in the state of overload or under load, the architecture can availably useHPSVFR to accomplish the VMs relocation.