Predicting Cloud Workloads based on Data Optimization Gradient Descent with Momentum (GDM) Approach

Cloud Computing has revolutionized computing off late with several domains and applications resorting to the cloud architecture. However effective task scheduling and load balancing is critical for cloud based servers. This is typically a very challenging task keeping in mind the fact that cloud workload is a parameter that depends on several other parameters. Moreover, due to the enormity of the data and its complexity, the use of machine learning or artificial intelligence based techniques is important for cloud workload estimation. Forecasting future workloads with high accuracy is especially challenging due to the randomness of the cloud workloads and also the non-deterministic nature of the governing or affecting parameters. Hence, due to the size and complexity of the data involved, finding regular patterns is a challenging task at hand. The present work proposes a back propagation based deep neural network architecture for cloud workload forecasting. The experiment uses the NASA cloud data set. The performance evaluation parameters have been chosen as mean absolute percentage error (MAPE) and regression. It has been found that the proposed system attains lesser mean absolute percentage error compared to previously existing technique [1].