A Resource Usage Prediction System Using Functional-Link and Genetic Algorithm Neural Network for Multivariate Cloud Metrics

Designing prediction-based auto-scaling systems for cloud computing is an attractive topic for scientists today. However, there are many barriers, which must be solved before applying these systems to practice. Some challenges include: improving accuracy for prediction models, finding a simple and effective forecast method instead of complex techniques, and processing multivariate resource metrics at the same time. So far, there are no existing proactive auto-scaling solutions for clouds that have addressed all those challenges. In this paper, we present a novel cloud resource usage prediction system using functional-link neural network (FLNN). We propose an improvement for the FLNN by exploiting genetic algorithm (GA) to train learning model in order to increase forecast effectiveness. To deal with multivariate input data, several mechanisms also are combined together to enable the ability of processing simultaneously different resource types in our system. This enables to discover implicit relationship among diverse metrics and based on that realistic scaling decisions can be made closer to reality. We use Google trace dataset to evaluate the proposed prediction system and data preprocessing mechanisms introduced in this work. The gained outcomes demonstrated that our system can work effectively under practical situations with good performance as compared with traditional techniques.