Using Structural Similarity to Predict Future Workload Behavior in the Cloud

Abstract

Predicting workload behavior, in response to changes in allocated resources, is a critical part of effective resource management in the cloud. This paper presents a novel approach to predicting the future behavior of a reference workload based on a nearest neighbor similarity search in euclidean space. The proposed approach involves identifying a similar workload candidate, predicting the future behavior of the reference based on the candidate and then validating the prediction using a statistical hypothesis test. Finally, decision rules are generated that specify the required conditions for a successful prediction.