Study on operating system architectures for supporting application-specific resource management

我们研究了特定于应用程序的资源管理的操作系统体系结构。传统上，资源管理策略是在操作系统内核中实现的。本研究探索新的操作系统对用户级资源管理的支持1。我们设计并实现了一种灵活的高效轻量级线程机制，该机制在用户级与操作系统内核协同实现。由于大多数线程操作是在用户级别执行的，因此可以通过使用特定于应用程序的策略来高效地执行线程。其显著特点是内核和用户级调度器的一致集成，以及高效的远程过程调用机制，从而消除了额外的用户-内核交互。我们为业务开发提出了一个有效的应用框架。它提供了基于线程迁移的进程间通信。进程间通信采用基本优先级继承协议。该框架简化了微内核与用户级服务之间的交互，保证了实时响应。设计了一种新的联合挂载路径名评估算法。与传统方法不同，该算法将分层的目录树一直合并到一个名称空间中；每个分层的目录可以具有不同的目录树结构；并且不仅允许修改前层的文件系统，也允许修改后面的层的文件系统。我们介绍了用于评估由联合挂载组成的分层名称空间中的路径名的主要规则，并在一个实际的操作系统中实现了这些规则。

We investigated operating system architectures for application specific resource management. Traditionally, resource management policies have been implemented inside the operating system kernel. This study explores new operating system support for user-level resource management.1. We designed and implemented a flexible mechanism for efficient lightweight threads which is implemented at the user-level in cooperation with the operating system kernel. Since most thread operations are performed at the user-level, threads can be executed efficiently by using application specific policies. The distinguished features are the coherent integration of kernel and user-level schedulers and the efficient remote procedure call mechanism which eliminates extra user-kernel interactions.2. We proposed an effective application framework for service developments. It provides the peocedural interprocess communication based on the thread migration. Basic priority inheritance protocol is applied to the interprosess communication. The proposed framework simplifies the interaction between the micro-kernel and user-level services, and ensures real-time response.3. We designed a new pathname evaluation algorithm for union mount. Unlike the conventional method, the proposed algorithm merges all the way down the layred directory trees into a name space ; each of the layred directories can have different directory tree structures ; and modifications are allowed on not only the file-systems in the front layr but also the file-systems in the layrs behind. We introduced principal rules to be used for evaluating pathnames in a hierarchcal name space composed of union mounts, and implemented these rules in a practical operating system.