The Research Of Virtual Machine Live Migration And Hot-spots Elimination Strategy Based On Resource Load Balancing

In today's era of big data,cloud computing as a service model for IT departments,research departments,the business sector to provide dynamically scalable services.VM enables logic isolation physical resources across systems to provide users with the service and support dynamic migration,solve the user at different times elasticity of demand for virtual resources.However,due to cloud platform has huge resources,uneven distribution and the complicated structure,coupled with the random of user tasks and share resources,can easily cause the entire system resource load imbalance and a virtual machine frequent phenomenon of migration hots-pots.At present domestic and foreign scholars in the virtual machine live migration policy mostly only consider the resource load balancing,and did not consider the correlation and anti-affinity binding migration services.Eliminate hot spots for virtual machine research,we have only at how to meet the mission requirements of the user premise Virtual machine migration policy,and did not consider the system of integrated resource load capacity.In this paper,load balancing of virtual machine CPU,memory,network bandwidth,and I/O,four-dimensional resource as a starting point,were presented live migration strategy based on resources load balancing of virtual machines based on performance prediction algorithm for virtual machine hot elimination strategies,and through experiments demonstrate the proposed algorithm analysis.In this article,the main work has the following three aspects:(1)For the current virtual machine live migration policy focus only resource load without considering the relevance of the migration service,we consider multidimensional resource load balancing,and the associated and anti-affinity binding in migration service portfolio,proposed a virtual machines dynamic migration strategy based on genetic algorithm.In the dynamic migration process,consider the relevance and anti-affinity binding request service and call volume queries and query dimensional constraints of resources.Compared with traditional genetic algorithm in the virtual machine live migration policy,combined with the algorithm thought the service portfolio,the request is a virtual machine migration as a collection by different migration service composition,and host capacity and query binding inquiry service.(2)For the problem of Eliminate hot spots for virtual machines,we consider the current usage of cloud platform resources,using resource prediction algorithm to predict the value of the nearest future time physical host.In the virtual machine live migration trigger selection strategies and objectives of physical hosts were calculated predicted value of resources,in order to make a good prediction of the performance of a physical resource node future time.In which the performance prediction algorithm using singular value decomposition theorem,the virtual machine migration process performance four-dimensional resource extraction,preservation of historical resources load value information for each physical host,and then find the next time the virtual machine to be migrated resource performance forecast estimates.(3)The experimental design was analysis carried out for the proposed algorithm.Experimental design is by increasing the number of physical machine data center,the number of virtual machine and binding conditions to detect the number of migration algorithm and the influence of the migration time.Result show that,as the host number increasing,number of migration and migration time also increases.To increase the number can increase the number of virtual machine migration and migration time.Inverse correlation constraint conditions to increase the number of migration and migration time no much impact but make the choice of host number increase.For the virtual machine live migration experiment data show that the hot spot eliminating strategy with the traditional average prediction algorithm and weighting value prediction algorithm,compared the performance prediction algorithm based on singular value decomposition theorem in the virtual machine migration has more accurate prediction results,can decrease the number of migration for the virtual machine as a whole,has a better forecast on the happening of the hot issues.