Concurrent reading and writing with replicated data objects

Abstract

Sharing the components of a data object among asynchronous processes is discussed in the context of the classic “Readers and Writers” problem. Distinct process groups identified as “Readers” and “Writers” contend for access to the data object with the provision that “Readers” may operate in a mutually concurrent way while “Writers” require exclusive access to the object. Two solutions are presented based on the use of replicated versions of the data object. The first algorithm solves a potential starvation problem persistent in previously reported solutions, while the second eliminates synchronization requirements between “Readers” and “Writers” by incorporating an optimized retry mechanism.