CNS: Core: Small: Consistent, Geo-Distributed Data Stores on the Public Cloud Using Erasure Coding

基于云的数据存储采用数据复制与算法相结合，这些算法允许一致、并发地访问数据，从而为其全球客户群提供较低的响应时间。擦除编码（EC）是复制的一种推广，在理论上，对于相同的容错能力，它具有更高的存储效率。然而，在实践中收获EC的地理分布的数据存储的好处提出了一些新的科学挑战。该项目的总体目标是通过全面了解基于EC的一致性地理分布商店的响应时间与成本权衡来应对这些挑战。研究将在以下四个方面进行：（i）开发EC技术和分布式算法，以满足称为最终和因果一致性的一致性标准，（ii）开发有效的、EC兼容的重新配置算法，其可证明地保持一致性，而不管对象配置的变化，（ii）开发性能和成本的分析模型，以及相关的优化框架，以及（iv）将开发的技术集成到Apache Cassandra中。拟议的研究将在公共云的背景下进行，其中DC间的延迟和定价信息随时可用，允许以公平的方式比较竞争方案。拟议的研究可能会降低托管在公共云上的地理分布式数据存储的成本，这些数据存储是多个应用程序的基本构建模块，包括协作编辑，社交媒体，金融交易，预订系统和多玩家游戏。验证计划涉及使用基于Apache Cassandra的原型进行的实验，该原型将成为开源。该原型将作为其他研究人员和工程师的测试平台，插入他们自己的算法并比较结果。这项研究具有跨学科的性质，并将补充一个教育计划，涉及调查文章的发展和课程整合。这个奖项反映了NSF的法定使命，并已被认为是值得通过使用基金会的智力价值和更广泛的影响审查标准进行评估的支持。

Cloud-based data stores employ data replication combined with algorithms that allow consistent, concurrent access to data to offer low response times to their global client-bases. Erasure coding (EC) is a generalization of replication known to be more storage efficient for the same fault tolerance in theory. However, reaping the benefits of EC for geo-distributed data stores in practice poses several new scientific challenges. The overarching goal of this project is to address these challenges by developing a comprehensive understanding of the response time vs. cost trade-off for EC-based consistent geo-distributed stores. The research will be conducted in the following four thrusts: (i) development of EC techniques and distributed algorithms to satisfy consistency criteria known as eventual and causal consistency, (ii) development of efficient, EC-compatible reconfiguration algorithms that provably maintain consistency despite changes in object configurations, (ii) development of analytical models of performance and cost, and associated optimization frameworks, and (iv) integration of developed techniques into Apache Cassandra. The proposed research will be conducted in the context of the public cloud, where inter-DC latencies and pricing information are readily available allowing competing schemes to be compared in a fair manner.The proposed research could lead to cost reductions for geo-distributed data stores hosted on public clouds, which are fundamental building blocks for several applications including collaborative editing, social media, financial transactions, reservation systems, and multi-player gaming. The validation plan involves experiments performed using an Apache Cassandra-based prototype that will be made open source. The prototype will serve as a testbed for other researchers and engineers to plug in their own algorithms and compare findings. The proposed research has a cross-disciplinary nature and will be supplemented with an education plan that involves development of survey articles and curricular integration.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.