A Load Balancing Method Based On Service Load Measurement And Forecast In Virtualized Environment

Network-based low-cost hardware cluster and grid computing systems are developing rapidly, which promotes the development and application of virtualization technology on the Internet. But there followed a problem of unbalanced resource utilization, and load balancing is a key technology to improve system resource utilization and parallel computing performance. Therefore, the study of load balancing has been extensively studied.Load balancing is inseparable from the monitoring technology.Currently the existing monitoring technology mainly stays in server level,physical node level and virtual machine (VM) level. The load balancing researches mainly focused on the load balance between physical nodes and physical servers. VM migration and service migration are the primary methods of load balancing, which balance load by migration rather than by forwarding requests to achieve load balancing. However, most of the current dynamic VM migration load balancing is based on single-target load balancing. In addition, VM migration will produce a pause in order to keep the memory state during the migration. Compared with service migration, VM migration is coarser, higher resource consumption, longer and more complex of migration process.In view of the shortcomings of VM migration people began to study on how to achieve load balancing through service monitoring and service migration. Service migration has the advantages of small migration scale,fast speed and low impact of migration on service quality. However, the service monitoring technology is not mature yet, the service impact on the VM load indicators is difficult to define and hard to collect effectively,and service migration often Luses a single service migration, which results in slow speed of load bala.ncing. In service migration, the threshold method is often used to decide the overload. If a VM is found overloaded,the service on this VM will be migrated inaccurately. 'The service that is to be migrated might be moved blindly to the target VM.in order to solve the problem of migration jitter caused by neglect of service load on virtual machine load, this paper proposes a load balancing method based on service load measurement and forecasting in virtualized environment. In this paper, the performance impact data of a. service to a VM is calculated first, and then the load formula of the service can be achieved through polynomial fitting, and after that the dual-way overload reporting mechanism of the VM can be realized by combining the service formula and the VM load forecasting. In the calculation center, the service load is calculated by using the service load formula. The services to be migrated are selected according to the objective function. Then, the optimal target algorithm is used to select an appropriate target VM for the services. Finally, the services are dynamically pushed to and deployed on the target VMs, so that load balancing is achieved. This paper also set up a Xen-based virtualized environment, and realized the automatic service migration load balancing system based on service load measurement and forecasting.