Dynamic allocation strategy of VM resources with fuzzy transfer learning method

Predicting virtual machine (VM) workload to realize dynamic allocation resources has always been a hot issue in research, most of the current resource prediction methods are based on different load resources to build prediction models, it is difficult to realize knowledge transfer between multi-task prediction models to complete multiple tasks’ prediction. This paper proposes a innovative method - dynamic resource allocation method based on fuzzy migration learning, which is based on the feature attributes of command line processes to predict multiple resource loads of VMs and realize dynamic allocation of VM resources. Firstly, Principal Components Analysis (PCA) algorithm is used to reduce the attributes dimension of command line process. Then, we apply fuzzy transfer learning, which is based on fuzzy neural network with the capability to deal with the uncertainty in transfer learning, to predict multiple resource loads of VMs with the strong regularities of command line processes, and dynamically configure the resources of VMs. In the experimental procedure, we take CPU and memory as examples, on the basis of CPU prediction model, the model parameters are transferred to memory prediction model obtaining good results. The implementation verified the effectiveness of the proposed method, achieves the aim of dynamic resource allocation of VMs, and improved the VMs’ performance.