Description:

tin is a database engine so tiny, its name had to be shortened. More specifically, tin focuses on storage, retrieval, subscription and transformation of sequential data. Like many other "NoSQL" solutions, it is not meant as a general-use RDBMS replacement but finds all its strength when handling data such as Facebook walls, Twitter feeds or any multi-dimensional data that is primarily served or ordered along a single dimension (or single set of dimensions). The approach at the core of tin was first tested on financial data, in the back-end delivery service behind Nasdaq's Market Replay --- where it was able to serve petabytes worth of stock prices daily without breaking a sweat (or the piggy bank).

This talk will cover:

the reasons that led to the conception of such a tool; how tin and other "text & filesystem" DBMS can be architected; how to reduce some data conundrums to 1-dimensional problems; performance gains of such approaches (the usual charts & graphs pr0n) and as a bonus (if time allows): "lens-based" transformation systems like the one used inside tin.