SPX: Collaborative Research: Distributed Database Management with Logical Leases and Hardware Transactional Memory

数据库系统是维护大量数据的关键应用程序的基础。自从十年前单处理器性能趋于稳定以来，增加处理器或服务器的数量已经成为提高分布式数据库管理系统性能的唯一可行方法。由于大量并行任务之间的复杂协调，这些系统的可扩展性是一个艰巨的挑战-这个项目试图解决的问题。大多数现有的数据库系统使用传统的物理时间来确定并行任务之间的顺序。这些系统需要管理分布式锁，这会导致阻塞和计算开销。其他系统使用逻辑时间（可以被认为是顺序中的位置）来消除锁定，但需要集中生成顺序，这是一个严重的可扩展性瓶颈。这个项目打破了物理时间的抽象，取而代之的是一个新的时间定义，包括逻辑和物理方面。“生理”时间，为了便于发音而被称为生理时间，使用逻辑时间戳来排序事件，然后使用物理时间打破联系。这使得新的依赖性避免方法，以提高系统的性能和可扩展性。本计画将生理时间应用于一个分散式资料库系统中的三个元件。(1)在硬件层面，将建立一种新的硬件事务存储器（HTM）机制，它允许多核处理器的缓存中更有效的数据移动。(2)设计并实现了一种新的分布式并发控制协议，以协调分布式数据库中的大量并行任务。(3)一个有效的并行索引数据结构将被提出用于多核和分布式数据库。该项目的所有三个部分都将在硬件/软件测试平台上进行原型设计和部署。该奖项反映了NSF的法定使命，并被认为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来支持。

Database systems are the foundation of critical applications that maintain large amounts of data. Since single-processor performance plateaued a decade ago, increasing the number of processors or servers has become the only viable way of improving performance in distributed database management systems. Scalability is a daunting challenge in these systems due to the complex coordination among the large number of parallel tasks---a problem that this project seeks to solve. Most existing database systems determine the order among parallel tasks using conventional physical time. These systems require managing distributed locks, which leads to blocking and computation overhead. Other systems use logical time, which can be thought of as position in an order, to eliminate locking, but require centralized generation of the ordering, which is a serious scalability bottleneck as core count increases. This project breaks the abstraction of physical time and replaces it with a new definition of time that incorporates both logical and physical aspects. "Physiological" time, termed physiological time for ease of pronunciation, uses logical timestamps to order events and then breaks ties using physical time. This enables novel dependency-avoiding approaches to improving system performance and scalability. This project applies physiological time to three components in a distributed database system. (1) At hardware level, a new hardware transaction memory (HTM) mechanism will be built, which allows more effective data movement in multi-core processors' caches. (2) A new distributed concurrency control protocol will be designed and implemented to coordinate large numbers of parallel tasks in a distributed database. (3) An efficient parallel indexing data structure will be proposed for both multi-core and distributed databases. All three parts of the project will be prototyped and deployed in hardware/software testbeds.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.