2-tier vs. 3-tier Architectures for Data Processing Software

The rise of data-centric computing, NoSQL and newSQL databases with powerful scripting capabilities, popularity of REST API raise a question: is it feasible to serve clients directly from the DB, with REST API server residing inside the database? What would be the balance between data processing, application, and presentation logic for such a scenario on a server side and on a client side? In this paper we will compare 2 real-world implementations of the commercial Luxms BI analytical platform based on 2-tier and on 3-tier architecture. Our research shows that despite popularity of 3-tier architectures, in-database application server approach delivers better performance in both throughput and latency in analytical client-server application development.