Engineering Distributed Systems that We Can Trust (and Also Run)

Abstract

The interest in formal methods and verification of correctness-critical distributed systems is on the rise in the past few years. But what are the gains from proving statements about software in full mathematical rigour? Do they justify the high cost of verification? And how far can we extend our trust in formal methods when talking about realistic distributed systems and their client programs? This talk is in three parts. First, I will provide an overview of the state of the art in machine-assisted reasoning about distributed consensus protocols, their implementations, and applications. Next, I will discuss the trade-offs that have to be made in order to enable mechanised proofs about runnable systems code, as well as implications of the assumptions made to describe the real-world execution environments. Lastly, I will focus on the ongoing work propelled by the programming languages community towards engineering modular proofs about distributed protocols-a way to build correct-by-construction composite systems from verified reusable components.