A Hierarchical Framework with Consistency Trade-off Strategies for Big Data Management
Yingyi Yang, Yi You, Bochuan Gu · 2017
Geo-replicated cloud storage provides good scalability, availability and fault-tolerance for managing big data of high-volume, velocity and variety nature. However, the prickly trade-off between consistency, cost and response time, which is brought in by geo-replicated cloud storage, poses great challenges to big data management. The goal of this work is to allow geo-replicated cloud storage used in big data management to dynamically switch to an appropriate consistency level at runtime in the consideration of cost and performance constraints. In this paper, we present a hierarchical consistency framework which supports the implementations of a strong protocol and a range of consistency semantics. The framework adopts a set of consistency trade-off strategies at both the data level and the transaction level. Based on a probabilistic model, a cost balance formula and a new metric Consistency-ResponseTime Efficiency defined in the trade-off strategies provides a basis to dynamically switch consistency levels by using performance records collected at runtime. Our evaluation verifies the effectiveness of our hierarchical framework and trade-off strategies.