ABSTRACT
The cloud service providers serve their consumers through a large scale of resource pool. The operation of these resources consumes an enormous amount of energy and causes high carbon footprints. The modern data center must optimally use its resources to minimize the energy consumption, carbon footprints, and operational cost. The workload forecasting has been very useful in cloud management. In this chapter, two predictive frameworks using ensemble learning are discussed. The ensemble learning uses multiple base predictors and their opinion is considered in obtaining the final forecast. The frameworks use an ensemble learning machine as an expert. The first framework decomposes the workloads into three simpler components and every component is modeled using an expert. The final forecast calculation involves the incorporation of individual forecasts from experts. The second framework also uses a set of base experts to forecast to model the workloads. In this framework, the base experts have different architectures and each network models the workloads independently. These forecasts are further weighted using a voting engine. The framework learns the weights for every base expert's forecasts using a blackhole algorithm. The performance of these frameworks is assessed on a variety of experiments.
