Build Stuff'15 Lithuania

A walk through of key pieces of a working production architecture that performs real-time analytics and visualization on 113 million data points per second on a single desktop class workstation. This was achieved through a combination of message oriented processing and unsafe data structures in key locations.
We will review how we mixed managed code across the majority of the application with unsafe data structures in key algorithmic location giving the best of both world.
We will review the details of simple custom memory management used in the allocation unsafe data without leaks or GC thrashing and some of the particulars of the general algorithmic approaches leveraging data locality and pointer operations.
Finally we will review the message based data processing pipeline that routes the processing through the system.