Specification and implementation of resilient, atomic data types

Abstract

A major issue in many applications is how to preserve the consistency of data in the presence of concurrency and hardware failures. We suggest addressing this problem by implementing applications in terms of abstract data types with two properties: Their objects are atomic (they provide serializability and recoverability for activities using them) and resilient (they survive hardware failures with acceptably high probability). We define what it means for abstract data types to be atomic and resilient. We also discuss issues that arise in implementing such types, and describe a particular linguistic mechanism provided in the Argus programming language.