Enhancing Data Consistency via a Context-Aware Dynamic Adaptive Model

Abstract

This paper introduces a new model for managing data consistency in large-scale, geo-distributed storage systems, pivoting on the concept of dynamic adaptive consistency. Recognizing the challenges in balancing consistency and availability in such systems, we propose an innovative, context-aware model that categorizes operations into "consistent blocs." This categorization allows for a more granular and efficient management of consistency levels, representing a notable improvement over adaptive approaches that employ a uniform consistency model across all operations or apply a single consistency level to each operation independently. Our model dynamically adapts these blocs' consistency levels in response to real-time changes in the operational context; this can include, for example, variations in network latency, data access patterns, and workload intensity, ensuring optimal data consistency tailored to current conditions. Our approach extends traditional adaptive consistency models by introducing more flexibility. A middleware architecture achieves this goal by introducing an Adaptation Manager that dynamically adjusts consistency levels. We implement this model and evaluate its performance using the YCSB benchmark on a Cassandra cluster. Our results reveal significant flexibility in expressing users' requirements and prompt responsiveness in the dynamic adaptation of the policy. Our proposition holds significant benefits for applications where rapid adaptation to context is crucial.