CAREER: Transparent Support for Efficient, Wait-Free Transactions in Shared-Memory Multiprocessors

在科学计算和数值计算中，多处理器计算机系统的使用越来越普遍。在为多处理器系统开发应用程序时，在协调对共享数据的并发访问方面出现了新的挑战。这通常是通过锁定实现的，这通常会限制并行性，并且还会导致性能瓶颈，因为延迟的进程可能会导致其他进程等待。这项工作的目的是通过使用透明的、无等待的用户提供的事务实现来消除多处理器应用程序中的锁定，从而确保延迟的进程永远不会导致其他进程等待。在这个项目中，将设计无等待实现，自动解决并发事务之间的冲突，并利用无冲突事务之间的并行性。通过仿真和应用对其性能进行研究。最后，将在多个系统上开发应用程序代码的无缝接口。该项目的教育计划侧重于将并发编程的研究与教育相结合。本研究将为在多处理器系统中开发高效的应用程序提供新的工具。这些工具将在未来的并发编程课程中用到。＊＊＊

The use of multiprocessor computer systems is increasingly common in scientific and numeric computing. In developing applications for multiprocessor systems, new challenges arise in coordinating concurrent accesses to shared data. This is traditionally achieved through locking, which often restricts parallelism, and can also cause performance bottlenecks because delayed processes can cause others to wait. This work aims to eliminate locking from multiprocessor applications through the use of transparent, wait-free implementations of user-supplied transactions, which ensure that delayed processes never cause others to wait. In this project, wait-free implementations will be designed that automatically resolve conflicts between concurrent transactions, and exploit parallelism between nonconflicting transactions. Their performance will be studied through simulation and applications. Finally, a seamless interface for application code will be developed on several systems. The education plan of this project focuses on integrating research and education in concurrent programming. This research will provide new tools for developing efficient applications in multiprocessor systems. These tools will be used in some of the future courses in concurrent programming. ***