Amazon Web Services runs some of the largest distributed systems in the world. AWS hosts data and applications for a myriad of organizations including large enterprise businesses, government agencies, researchers, and internet startups. The data gets bigger and bigger; the systems grow at astonishing rates, and the software supporting these systems gets increasingly complex. 

We take the availability and durability of these data very seriously. It is of the utmost importance to us that data is never lost or corrupted, is highly secure, and is highly available. At the same time we must continue to innovate, offer greater performance and functionality, and scale these systems while maintaining a sustainable operations profile.

To do all of this well means that the software must be bullet-proof when we clear it for deployment to production systems. Algorithm correctness becomes ever more critical. This talk will take a look at how testing strategies have evolved, and continue to evolve within AWS in order to achieve greater reliability at ever more massive scale. We'll start with the conventional methods, and then roll forward through time covering how and why we incorporated more powerful testing methodologies, ultimately leading us to the use of formal methods where TLA+ has become a cornerstone to our overall strategy.