Where: Outhouse Pub @ 7pm
Who: Miles Gould
Title: CRDTs and LVars
Abstract
---
The CAP theorem for distributed databases ("consistent, available,
partition-tolerant: pick two") and the growth of Big Data have led to a
renewed interest in relaxed standards of consistency for distributed
data. For many applications it's enough to guarantee eventual
consistency between nodes, but designing protocols to ensure this has
proved tricky and error-prone. A neat approach is to build eventual
consistency guarantees into the datatypes themselves using ideas from
the theory of semilattices: this leads to so-called Convergent
Replicated Data Types (CRDTs). These turn out to be closely related to
the notion of LVars, studied by Kuper et al in the context of concurrent
shared-memory Haskell programs.