A Transaction Model for Management for Replicated Data with Multiple Consistency Levels

Abstract

We present here a transaction model for managing replicated data with different consistency guarantees to make suitable trade-offs between data availability, scalability, and consistency for different data items. Data is organized in a hierarchy which is based on consistency levels and the associated transaction management protocols. The model presented here is based on imposing certain constraints on information flow across data layers in this hierarchy to ensure the desired consistency guarantees for data items at different levels. The transaction model proposed here simultaneously supports transactions with different consistency levels, which include ACID transactions for strong consistency, and weaker consistency models such as snapshot isolation (SI), causal snapshot isolation (CSI), CSI with commutative updates, and CSI with unsynchronized concurrent updates. The building block for this transaction model is the snapshot isolation model.