CAREER: Fluid Replication

分布式信息系统的性能正变得越来越不可预测。过去解决这个问题的方法要么只处理根源的一个子集，要么引入自己的新困难。该项目提出了流体复制——自动创建和放置副本以响应不断变化的客户需求和网络容量——作为解决这个问题的方法。这些副本保存在WayStations上，WayStations是基础设施服务节点，可以充当本地和访问用户的副本站点。本研究的目的是构建实现流体复制的四个关键机制：端到端网络性能评估器、定位WayStations的可扩展方案、WayStations与远程服务器之间的一致性管理技术，以及从一个副本站点到另一个副本站点的高效客户端迁移。这些机制的原型将在实验和部署的网络测试平台中使用标准工作负载进行评估。通过这些机制，流体复制可以自动将可用资源与提供的负载相匹配。这降低了管理成本，提高了资源利用率，并为最终用户提供了更好的性能。

The performance of distributed information systems is becoming increasingly unpredictable. Past approaches to this problem either handle only a subset of root causes, or introduce new difficulties of their own. This project proposes fluid replication --- the automatic creation and placement of replicas in response to changing client demands and network capacities --- as a solution to this problem. These replicas are held on WayStations, which are infrastructural service nodes that can act as replica sites for both local and visiting users. The objective of this research is to construct four key mechanisms to enable fluid replication: an estimator for end-to-end network performance, a scalable scheme to locate WayStations, techniques for consistency management between WayStations and remote servers, and efficient client migration from one replica site to another. Prototypes of these mechanisms will be evaluated with standard workloads in both experimental and deployed networking testbeds. With these mechanisms, fluid replication promises to automatically match available resources to offered load. This reduces cost of administration, increases utilization of resources, and delivers improved performance to end-users.