Resource Availability Forecasting for Federated Clouds

Abstract

Cloud federation has enabled organizations to adopt collaborative services for sharing data and workloads across various platforms. Induction of federation members may require some verifications and predictions related to capacity and capability of these members for compensating such types of workloads. However, the nature of federated services require stringent methods to keep track of dynamically forming resource clusters for forecasting their behavior. Recent literature has mostly focused on the applicability of forecasting algorithms based on static datasets with little or no applicability to real time scenarios. Proposed research has utilized a real world application of Clouds4Coordination (C4C) federation system. A resource forecasting strategy using two well-known algorithms, namely Autoregressive Integrated Moving Average (ARIMA) and Long Short-Term Memory (LSTM) has been proposed for collaborative clouds used in Architecture, Engineering and Construction (AEC) industry. The results have shown that no single algorithm is sufficient enough to deal with dynamic scenarios of cloud federation. Moreover, the selection of algorithm is highly dependent upon the type and duration of prediction required i.e. short term or long term as required by the user.