Automated Cloud Scaling with GRU, LSTM, VARMAx Predictions

The proliferation of cloud computing has resulted in a paradigm shift from traditional computing. As a result of this transition the cloud servers are overloaded due to resource constraints and inefficient job scheduling methods. An efficient job scheduling mechanism is of paramount requirement due to the continuous demand for these resources. The traditional algorithms performance may fail to maximize the performance matrix such as maskespan, deadline hit ratio and scheduling efficiency level. In this research a method is proposed to mitigate these constraints, which enables real time cloud resource scaling. In the proposed method the algorithms such as Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) neural networks are integrated. The resource demand can be forecasted by using Vector AutoRegressive Moving Average with eXogenous variables (V ARMAx). As a result, a better performance is achieved by predicting the resource requirement and resource scaling proactively by utilizing V ARMx. The outcome of this research shown an 8.5% boost, 2.9% enhancement in the deadline hit ratio and a 3.5% improvement in makespan.