Query co-planning for shared execution in key-value stores

Abstract

Large amounts of data are being stored and queried using different data models. For each of these models, there are specialized data stores which are then accessed concurrently by many different applications. For instance, key-value stores provide a simple data model of key and value pairs. Thus, the simplicity of their read and write interface. Additionally, they provide other operations such as full and range scans. However, along with its simplicity, key-value stores impose some limitations when trying to optimize data access. In this work, we study how to minimize the data movement when executing a large number of range queries on key-value stores. This is based on the observation that when accessing a common dataset, there is usually a (possibly large) overlap among queries accessing it. Thus, to accomplish this, we use shared-workload optimization techniques to execute a group of queries together. We analyze different data structures suitable for co-planning multiple range queries together in order to reduce the total amount of data transferred. Our results show that by co-planning a group of range queries we reduce the total execution time of a query workload