Scott has been writing software professionally for 20+ years. He's been entranced by performance and efficiency at least 15 of those 20+ years. He worked in distributed systems, search, Adwords, Adsense and ML at Google for 13 years. He worked 4.5 years at Snap, on monetization systems, performance, and advanced mobile telemetry systems. He is currently an engineer at Rockset, working on realtime analytics. Presenter of a production software talk at Google, Snap, UCLA, USC, Caltech and others over last ten years.
This presentation is about how Generic Programming Paradigm support in C++ and its portability has enabled us to achieve very good performance and accidentally create a powerful analysis instrument to observe programming technique performance and real "silicon" execution capabilities. We will start by showing how our Hash Tables are "adaptive". We mean by "adaptive" that there are template parameters to modify memory layouts, hashing options, structure organization, and -crucially- bit widths of substructures. Then, at runtime, a table can transform itself into another table with other parameters. We applied the Generic Programming Paradigm from the start, resulting in stunning configurabiity, […]