Abstract

Modern databases underlying large-scale Internet services guarantee immediate availability and tolerate network partitions at the expense of providing only weak forms of consistency, commonly dubbed eventual consistency. Even though the folklore notion of eventual consistency is very weak, in reality, the semantics and programming interfaces of systems implementing it can be very subtle. Thus, such systems can resolve conflicting updates using complex protocols, called replicated data types, and can allow varying kinds of anomalous behaviour.

So far the semantics provided by eventually consistent systems has been poorly understood. I will present a framework for its formal specification that addresses this problem by generalising weak shared-memory models to arbitrary replicated data types. I will then describe our ongoing effort to develop effective reasoning principles for eventually consistent systems.

This is joint work with Sebastian Burckhardt (Microsoft Research), Hongseok Yang (University of Oxford) and Marek Zawirski (UPMC & INRIA ). Part of it will appear in POPL ’14.