Self-Partitioning Cloud Datastore for Scalable Transaction Processing Proposal
The authors are designing a new cloud data storage system that will be highly scalable and yet still be able to provide strong data consistency guarantees. Their system extends existing cloud key/value stores with modular components for transaction management and access-based partitioning. They plan to utilize graph theory, clustering algorithms, and Markov chaining to co-locate related data for efficient transaction processing and to assign partitions to optimal locations in the server clusters. Through continuous analysis of execution logs, their system should be able to recognize data access patterns and use that knowledge to partition data within the cluster to better fit within those patterns.