Workload Prediction with Regression for Over and Under Provisioning Problems in Multi-agent Dynamic Resource Provisioning Framework

The infrastructure of cloud computing is distributed environment, where resources, such as CPU, storage, and network bandwidth, are shared. The cloud computing system can have the over and under provision problems. Recently, the problems have been handled using the multi-agent dynamic resource provisioning framework, where Machine Learning algorithm is used for workload prediction. We empirically investigated the effectiveness of regression models for the over and under provisioning on cloud system. We found that while linear regression, polynomial regression, and support vector machine (SVM) performed equally well in term of workload prediction. However, polynomial regression was able to distribute the load among the hosts better. This resulting in the lower total energy consumption. All regression model requires little time for parameter tuning. As such, we suggest that the model's parameters should be readjusted as needed.