Study of different forecasting models on Google cluster trace

Workload prediction in cloud system is an important task and it helps in efficient resource allocation by minimizing cost and thus maximizing the profit. In this paper we analyze a large scale production workload trace (version 2) which is recently made publicly available by Google. The main objective of our research is to design and compare different forecasting models. We develop models through Adaptive Neuro-Fuzzy Inference System (ANFIS), Non-linear Autoregressive Network with Exogenous inputs (NARX), and Autoregressive Integrated Moving Average (ARIMA). Finally, we compare these three prediction models to find out the best one. Performance of forecasting techniques is measured by two popular statistical metrics, i.e., Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE). The experimental result demonstrates that NARX model outperforms other models, e.g., ANFIS and ARIMA.