Title: Lara: A Key-Value Algebra underlying Arrays and Relations

Abstract: Data processing systems roughly group into families such as relational, array, graph, and key-value. Many data processing tasks exceed the capabilities of any one family, require data stored across families, or run faster when partitioned onto multiple families. Discovering ways to execute computation among multiple available systems, let alone discovering an optimal execution plan, is challenging given semantic differences between disparate families of systems. In this paper we introduce a new algebra, Lara, which underlies and unifies algebras representing the families above in order to facilitate translation between systems. We describe the operations and objects of Lara---union, join, and ext on associative tables---and show her properties and equivalences to other algebras. Multi-system optimization has a bright future, in which we proffer Lara for the role of universal connector.