Actor Database Systems: A Manifesto

Interactive data-intensive applications are becoming ever more pervasive in domains such as finance, web applications, mobile computing, and Internet of Things. Typically, these applications are architected to utilize a data tier for persistence. At one extreme, the data tier is a simple key-value storage service, and the application code is concentrated in the middle tier. While this design provides for programmability at the middle tier, it forces applications to forego classic data management functionality, such as declarative querying and transactions. At the other extreme, the application code can be colocated in the data tier itself using stored procedures in a database system. While providing rich data management functionality natively, the resulting lack of modularity and state encapsulation creates software engineering challenges, such as difficulty in isolation of bugs and failures or complexity in managing source code dependencies. In addition, this monolithic architectural style makes it harder to scale the application with growing request volumes and data sizes. In this paper, we advocate a new database system paradigm bringing to developers the benefits of these two extremes, while avoiding their pitfalls. To provide modularity and reasoning on scalability, we argue that data tiers should leverage the actor abstraction; at the same time, these actor-based data tiers should offer database system features to reduce bugs and programming effort involved in state manipulation. Towards this aim, we present a vision for actor database systems. We analyze current trends justifying the emergence of this abstraction and discuss a set of features for these new systems. To illustrate the usefulness of the proposed feature set, we present a detailed case study inspired by a smart supermarket application with self-checkout.