Design And Implementation Of Load Balancing Mechanism Based On KVM Clusters

At present,cloud computing is rapidly developing and has become a hot topic in the academic and business circles.However,the rapid development of cloud computing also raises a series of issues.One of the critical issues is the unbalanced load in cloud clusters.Usually,a cluster consists of the physical servers from different manufacturers.Such heterogeneous physical servers also differ in the provision of different tasks,and there are a larger number of users' demands,thus the unbalanced load problem occurs.Then the traditional resource scheduling only considers the load characteristics of virtual machines,but ignores the types of virtual machines.While different types of virtual machines correspond to different resource requirements.which leads that the same physical machine for computation,memory,storage,network.etc is not balanced,which leads to the low utilization of the cluster resources.Concerning the above aspects,in this work,a load balancing mechanism is proposed based on KVM clusters.In the proposed mechanism,physical machines are classified into two categories:computing nodes and management nodes.Computing nodes are application providers while the management nodes are the core module of the load balancing mechanism,which consists of the load evaluation module,the elastic module,the core algorithm module and the elastic adjustment module.The load evaluation module gets the physical server's load weights based on the dynamic load balancing algorithm.The elastic module dynamically adjusts the scale of the KVM cluster according to the collected server's load weights.The core module is responsible to find the best matching of servers and the virtual machines by ant colony algorithm and simulated annealing optimization algorithm.Then the elastic adjustment module implements the migration of virtual machines and reevaluation of the entire cluster.In order to measure the load of each server,a dynamic load balancing algorithm is proposed in this work.It monitors and collects the run-time status of each virtual machine and each physical server,e.g.CPU utilization,memory utilization,10 occupancy rate and network bandwidth occupation rate.According to the capability and load information of each server in the data center.the load weight of each physical server is obtained.Then according to the load weight,it ensures the elastic expansion of the system,decides the time for virtual machine migrations and selects appropriate physical server so that the system load is balanced and the system availability and reliability are improved.To solve the matching problem between virtual machines and servers,an ant colony optimization and an simulated annealing algorithm(ACOSA)are proposed in this paper to find the best match between virtual machines and physical servers.Then the elastic adjustment module implements the migration of virtual machines and reevaluation of the system in order to ensure the load balance of the data center and stability,fluency of virtual machine.Finally,the proposed mechanism is evaluated via a CloudSim platform.The results show that the algorithm proposed in this paper is feasible.The proposed scheme is also evaluated on a practical KVM cluster.And the experiments show that the load balancing mechanism can effectively improve the resource utilization in the cluster and balance the work load among different physical machines.