Algorithms for practical distributed agreement

Agreement algorithms form the underpinning of many reliable services; they provide a consistent state among several machines despite asynchrony, crash failures, and malicious attacks. In this talk, Naama will argue that it is important to design agreement algorithms to optimize for both best-case and worst-case performance.

She will discuss a theoretical model that captures the behavior and costs of data-center networks on which agreement algorithms are often run, and will show recent results that provably improve the running time of agreement algorithms in both best- and worst-case executions.