On practical implications of trading ACID for CAP in the big data transformation of enterprise applications

Abstract

Transactional enterprise applications today depend on the ACID property of the underlying database. ACID is a valuable model to reason about concurrency and consistency constraints in the application requirements. The new class of NoSQL databases that is used at the foundation of many big data architectures drops the ACID properties and especially gives up strong consistency for eventual consistency. The common justification for this change is the CAP- theorem, although general scalability concerns are also noted. In this paper we summarize the arguments for dropping ACID in favour of a weaker consistency model for big data architectures. We then show the implications of the weaker consistency for the business customer and his requirements when a transactional Java EE application is transformed to a big data architecture.