CSR: Small: Collaborative Research: Improving Dependability of Multithreaded Distributed Programs

几乎所有现代处理器现在都包含多核，并且几乎所有现代计算机系统都包含多个处理器。因此，未来的软件可能既是多线程的(其中进程的线程使用共享内存进行通信)又是分布式的(其中系统中的进程使用消息进行通信)。确保程序在所有可能的情况下都能正常运行是一项非常困难的任务。大多数现实世界的程序包含大量的组件，这使得它们的正式验证是不可行的。在程序部署前进行测试和调试的有效工具是必不可少的。即使在广泛的测试和调试之后，错误仍然存在，特别是那些在极少数情况下出现的错误。在运行时监视程序并可能控制其执行以避免不良状态是容忍残留软件错误的重要方法。在这个项目中，我们致力于开发一种监控、分析和控制多线程分布式计算的理论和算法。具体地说，我们正在开发(I)一个统一的框架，用于在多核分布式系统中对由消息、锁和其他同步原语(例如，等待/通知)产生的同步进行建模，(Ii)使用切片和其他方法来检测和控制以时态逻辑公式表示的谓词的离线和在线算法，以及(Iii)用于跟踪事件之间的依赖关系的可扩展方法。除了多核计算，该工作还在其他领域得到应用，包括云计算、分布式数据库、恢复、副本一致性和资源管理。我们还在开发可在课程中使用的教育工具，以增强使用多核分布式系统的学生的学习体验。

Nearly all modern processors now contain multiple cores and almost all modern computer systems contain multiple processors. Thus future software is likely to be both multithreaded (in which threads of a process communicate using shared memory) and distributed (in which processes in a system communicate using messages). Ensuring that a program works correctly under all possible scenarios is a very difficult task. Most real-world programs contain a large number of components which makes their formal verification infeasible. Effective tools for testing and debugging programs prior to their deployment are indispensable. Bugs persist even after extensive testing and debugging especially those that manifest under rare circumstances. Monitoring programs at runtime and possibly controlling their execution to avoid bad states is an important way to tolerate residual software bugs. In this project, we are working on developing a theory and algorithms for monitoring, analyzing and controlling a multithreaded distributed computation. Specifically, we are developing (i) a unifying framework for modeling synchronization in multicore distributed systems resulting from messages, locks and other synchronization primitives (e.g., wait/notify), (ii) offline and online algorithms for detecting and controlling predicates, expressed as temporal logic formulas, using slicing and other approaches, and (iii) scalable approaches for tracking dependency among events.Besides multicore computing, the work has applications in a variety of other areas including cloud computing, distributed databases, recovery, replica consistency and resource management. We are also developing educational tools that can be used in courses to enhance the learning experience of students working with multicore distributed systems.