Presidential Faculty Fellows

该项目开发了一个名为SPIN的操作系统，它以安全、可预测和高效的方式为物理资源上的应用程序提供细粒度控制。SPIN系统是围绕可扩展的微内核体系结构构建的。系统是可扩展的，因为应用程序可以动态地安装受限制的程序序列，这些程序序列在内核中执行，以响应一组预定义的事件，从而实现灵活和快速的响应。例如，一个应用程序可以安装一个调度过滤器，每当该程序的地址空间中的线程被抢占时，该过滤器就会运行。事件过滤的概念可以应用于各种各样的事件和服务，包括虚拟内存、调度、网络通信、本地进程间通信和磁盘管理。通过将系统资源管理视为跨应用程序重新分配物理资源的过程，并将每次重新分配解释为一个独立的事件，SPIN为应用程序反馈定义了关键控制点，从而启用了动态资源分配策略。通过将应用程序定义的控制序列与这些控制点相关联，SPIN在每个事件中强制执行广泛的策略，从而使单个操作系统能够跨许多类型的应用程序良好地工作。通过在内核中执行控制序列，可以以非常低的开销实现动态控制。

The project develops an operating system called SPIN that provides fine-grained control to applications over physical resources in ways that are safe, predictable and efficient. The SPIN system is structured around an extensible microkernel architecture. The system is extensible in the sense that application programs can dynamically install restricted program sequences that execute in the kernel in response to a set of pre-defined events allowing for flexible and rapid response. For example, an application program could install a scheduling filter that runs each time a thread from that program's address space is preempted. The notion of event filtering can be applied to a wide class of events and services, including virtual memory, scheduling, network communication, local interprocess communication and disk management. By viewing system resource management as a process by which physical resources are reassigned across applications, and interpreting each reassignment as a discreet event, SPIN defines critical control points for application feedback, enabling dynamic resource allocation policies. By associating application-defined control sequences with these control points, SPIN enforces a wide range of policies at each event, enabling a single operating system to work well across many types of applications. By executing the control sequences within the kernel, the dynamic control is achieved with very low overhead.