Resource prediction based on program granularity combined with data purification

Resource prediction is the basis of resource allocation and task scheduling in cloud platform. Different from previous research which takes a server as the research object, we mainly study resource prediction based on program granularity and provide a resource prediction model. A program will cost Central Processing Unit (CPU), memory and I/O when it is executing. To predict these three kinds of resources, the Static Data of Servers (CPU main frequency, the number of CPU cores, etc.), the Static Data of Programs (Time Complexity, Space Complexity, etc.), and the Dynamic Data of Programs (CPU utilization, memory utilization, I/O) are collected. PauTa Criterion is used to purify the dataset in order to ensure the accuracy of resource prediction. Then a Back Propagation Neural Network (BPNN) predicts the resource. The momentum factor is added in a BPNN to improve the rate of convergence. In our experiments, different datasets are used to test the accuracy of our model, and the results show that the resource prediction model has a good performance.