The group paradigm for concurrency control

Abstract
We propose a paradigm for developing, describing and proving the correctness of concurrency control protocols for replicated databases in the presence of failures or communication restrictions. Our approach is to hierarchically divide the problem of achieving one-copy serializability by introducing the notion of a “group” that is a higher level of abstraction than transactions. Instead of dealing with the overall problem of serializing all transactions, our paradigm divides the problem into two simpler ones. (1) A local policy for each group that ensures a total order of all transactions in that group. (2) A global policy that ensures a correct serialization of all groups. We use the paradigm to demonstrate the similarities between several concurrency control protocols by comparing the way they achieve correctness.

This publication has 12 references indexed in Scilit: