Dynamic Workload-Aware Elastic Scale-Out in Cloud Data Stores

NoSQL databases store a huge amount of data generated by modern web applications. To improve scalability, a database is partitioned and distributed among the different nodes called as a scale out. However, this scale out feature of the NoSQL database is oblivious to the data access pattern of the web applications, which results in poorly distributed data across all the nodes. Therefore, the cost required for the execution of the query is increased. This paper describes the partition placement strategy, which will place data partitions to the available domains in the Amazon SimpleDB according to the data access pattern of web applications, which leads to an increase in throughput by some percentage. We present the workload-aware elasticity algorithm, which will not only add and remove the domain as per the load, but also places the partitions as per the data access pattern. We have validated the workload-aware elasticity and load distribution algorithm through experimentation over a cloud data store such as Amazon SimpleDB running in the Amazon Cloud. The throughput of the load distribution algorithm is predicted using the regression and the multiple perceptron model. Key–Words: partition placement, workload-aware elasticity, data partitioning, database scalability, placement strategy .