Edinburgh Napier University

  Cloud Computing has emerged as a low cost anywhere anytime computing paradigm. Given the energy consumption characteristics of the Cloud resources, service providers are under immense pressure to reduce the energy implications of the datacentres. Forecasting the anticipated future workloads would help the service providers to achieve an optimum energy-efficient scaling of the datacentre resources in accordance with the incoming workloads. But the extreme dynamicity of both the users and their workloads impose several challenges in accurately predicting their future behavioural trend. This paper proposes a novel prediction model named InOt-RePCoN (Influential Outlier Restrained Prediction with Confidence Optimisation), aimed at a tri-fold forecast for predicting the expected number of job submissions, session duration for users, and also the job submission interval for the incoming workloads. Our proposed framework exploits autoregressive integrated moving average (ARIMA) technique integrated with a confidence optimiser for prediction and achieves reliable level of accuracy in predicting the user behaviours by the way of exploiting the inherent periodicity and predictability of every individual jobs of every single users. Performance evaluations conducted on a real-world Cloud trace logs reveal that the proposed prediction model outperforms the existing prediction models based on simple auto-regression, simple ARIMA and co-clustering time-series techniques in terms of the achieved prediction accuracy.