CSR: Small: Practical Geo-Replicated Storage for Web Applications

点击翻译按钮获取中文摘要

There is increasing demand to deploy web applications across multiple datacenters to improve fault tolerance and reduce network delays tousers. As application servers need to share data, theirconstruction can be significantly simplified if a storage backend exists forapplications to seamlessly access replicated data at different data centers.A geo-replicated storage system faces the unpleasant tradeoff of consistencyvs. performance because of large inter-data-center communication delay. Thisproposal investigates novel consistency and programming models forgeo-replicated storage that are easy-to-use and can achieve good performance.In particular, this project proposes parallel snapshot isolation, a novel consistency model that enables efficient implementation with minimalcoordination across data centers. Parallel snapshot isolation provides muchstronger guarantees than existing weak consistency models by disallowingwrite-write conflicts and preserving the causality of operations. The PIbuilds Walter, a geo-replicated transactional key-value store that guaranteesparallel snapshot isolation. A number of common web applications are writtenon top of Walter and evaluations demonstrate that the new consistency model iseasy to program for and enables applications to achieve high performance. Thesuccess of this project will bring much improvement to the state-of-art inwriting scalable and fault tolerant web applications.