EAGER: Rapid Updates and Snapshot-Based Queries Using Multicore Processors

诸如交通监控、移动用户管理和传感器网络等应用程序需要处理大量更新，同时支持在线分析查询。有了大量的RAM，单个机器就有可能管理数以亿计的项目。使用多个硬件线程（现代商用多核芯片上多达64个），可以并发处理许多操作。同时处理查询和更新可能会导致干扰。查询需要看到一致的数据库状态，这意味着至少在某些时候，更新需要等待查询完成。为了解决这个问题，研究了各种解决方案，在这些解决方案中，在不同的时间点拍摄数据库的ram驻留快照。分析查询在快照上操作，消除了干扰，但允许答案稍微过时。目前正在开发和研究几种不同的快照创建方法，其目标是能够快速创建快照(例如：（在几分之一秒内），同时最大限度地减少更新处理的开销。这些问题既适用于传统的服务器机器，也适用于多核移动设备。通过在用户的移动设备上保持个性化的最新数据，可以支持广泛的潜在新应用程序，同时避免广泛传播一个人的位置所带来的隐私问题。研究的重点是如何有效地利用现代机器上的许多处理核心，包括传统和移动设备。主要目标是允许性能随着新一代硬件中可用的额外核心而扩展。这个项目的网站可以在http://www.cs.columbia.edu/~kar/snapshot.html找到

Applications such as traffic monitoring, mobile user management, and sensor networks need to process large volumes of updates while supporting on-line analytic queries. With large amounts of RAM, single machines are potentially able to manage hundreds of millions of items. With multiple hardware threads, as many as 64 on modern commodity multicore chips, many operations can be processed concurrently.Processing queries and updates concurrently can cause interference. Queries need to see a consistent database state, meaning that at least some of the time, updates will need to wait for queries to complete. To address this problem, a variety of solutions are explored in which a RAM-resident snapshot of the database is taken at various points in time. Analytic queries operate over the snapshot, eliminating interference, but allowing answers to be slightly out of date. Several different snapshot creation methods are being developed and studied, with the goal of being able to create snapshots rapidl(e.g., in fractions of a second) while minimizing the overhead on update processing.These problems are studied both for traditional server machines, as well as for multicore mobile devices. By keeping personalized, up to date data on a user's mobile device, a wide range of potential new applications could be supported while avoiding the privacy concerns of widely distributing one's location. The research focus is on how to efficiently utilize the many processing cores available on modern machines, both traditional and mobile devices. A primary goal is to allow performance to scale as additional cores become available in newer generations of hardware.The website for this project can be found at http://www.cs.columbia.edu/~kar/snapshot.html