On prediction of resource consumption of service requests in cloud environments

The functional efficiency of cloud solutions depends on their responsive allocation of processing resources to individual tasks according to their predefined completion time. An optimum usage of resources requires an accurate prediction of resource consumption. In this paper we introduce a novel method for predicting the resource consumption of processing requests in multiple consumption classes in a system consisting of a plentitude of processing units. We further enhance the prediction algorithm with different mechanisms to determine the request resource consumption of a variety of request classes in order to increase the precision of the prediction. We also show that the prediction algorithm is robust against abrupt and large variations of consumption and recovers itself rapidly.Index Terms: Resource consumption prediction, resource allocation, Machine Learning, adaptive algorithms, network softwarization, cloud computing, distributed systems.