Workload management for dynamic partitioning schemes in replicated databases

Recent advances on providing transactional support on the cloud rely on keeping databases properly partitioned in order to preserve their beloved high scalability features. However, the dynamic nature of cloud environments often leads to either inefficient partitioning schemes or unbalanced partitions, which prevents the resources from being utilized on an elastic fashion. This paper presents a load balancer that uses offline artificial intelligence techniques to come out with the optimal partitioning design and replication protocol for a cloud database providing transactional support. Performed experiments proof the feasibility of our approach and encourage practitioners to progress on this direction by exploring online and unsupervised machine learning techniques applied to this domain.