CSR: Small: Collaborative Research: Improving Latency in Geo-Replicated Storage by Relaxing Consistency Requirements

University Of Illinois At Urbana-Champaign, Urbana IL

Investigators

Abstract

Recent years have seen a tremendous growth in the popularity of cloud services. Distributed storage systems are a key component of the cloud computing revolution. Distributed storage systems are designed to achieve a suitable tradeoff between latency, data consistency and dependability. This project addresses development of new consistency models that take into account application characteristics and hybrid fault models. The ultimate goal is to improve our understanding of the trade-offs between consistency, latency, and dependability in distributed storage systems. The project includes two synergistic thrusts. First, the project explores consistency models that consider graph-based application characteristics. For example, social and trust graphs model user relation and interaction for many applications. The first thrust explores graph-based consistency models that are acceptable for the applications, and yet improve latency of client operations. Second, prior storage systems typically either only tolerate benign failures, or tolerate worst-case faults with high overhead. This project considers non-colluding faults as well as Byzantine faults, and investigates approaches to reduce overhead by relaxing some of the requirements imposed by prior systems. Proposed research is expected to improve the understanding of the impact of application characteristics and different types of faults on consistency-latency trade-off in distributed storage systems. Tolerance of more severe faults is likely to become important for future critical web-based services and applications. Additionally, with the tremendous growth in social and trust networks, mechanisms to improve their performance are of interest. This project helps achieve such improvements through the development of new consistency models. Additionally, the project provides research opportunities to undergraduate and graduate students, and a post-doctoral researcher. The data from the project is to be retained for at least two years after the completion of this project, and stored in local computers and online shared storage. The URL for the repository of the data, publications and software resulting from the project is https://sites.google.com/site/nsfgeoconsistency/. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.