Databases are typically utilized for storing data, but data is only useful if it is being turned into actionable insights. In this session we will explore an important adjacent-role of the database, to efficiently perform computations and create derived data.

Relational databases have been successful for several decades now because they provide a well defined computational framework, which is built on top of relational algebra. The relational algebra defines a set of operations that can be performed on the data.

However, times change and we are presently in a situation where relational databases do not easily scale to the amounts of data that businesses and organizations around the world are dealing with. NoSQL technology has emerged as an attempt to address this, but NoSQL loses the computational aspect. The majority of NoSQL databases are glorified key-value stores that support CRUD operations, but do not define a solid computational model.

Alternatively, MapReduce defines a computational framework to process large amounts of data, but has its own caveats. It encourages batch processing and full-scans of data, so the computation becomes detached from stored data.

It is our belief that the data management industry will move into the direction of consolidating storage, which will allow computing power and new databases to emerge; these will both scale to data storage needs as well as provide a solid online computational framework.

This session will teach you about a computing engine that is built into the database and allows arbitrary JavaScript-based code to be executed directly in the database, next to the data.

This setup provides a simple and powerful computational model that ensures data is accessed through efficient indices directly in the database, thus massively accelerating data manipulation operations.

Throughout this talk, we will present examples of how advanced computation can be done directly in the database, including an analysis of geospatial data, and the application of statistical model-like logistic regression to derive predictions.

Jurgis Orups

Co-Founder & CTO,
Clusterpoint

Jurgis co-founded Clusterpoint in 2006.

Prior to that, he spent 6-years managing a team building large-scale Internet crawler and search technology. He specializes in parallel computing, text retrieval algorithms and natural language processing.

Jurgis has an MSc in Computer Science from the University of Latvia. When not at work, he enjoys photography and mountain biking, and even keeps a training bike in the office.