ARTOS: adaptive resource scheduling for component-based soft real-time applications

实时系统大多需要预先了解所需的资源消耗情况。这与现代使用场景是矛盾的，在现代使用场景中，应用程序（app）被下载并安装在不同的系统上，并且彼此并发使用。如果应用程序使用已经存在的组件，则几乎不可能准确地说明所需的资源。因此，我们的目标是一个能够支持具有软实时需求的应用程序的平台，例如多媒体应用程序。这些可以动态地安装和使用。考虑到最终用户的偏好，他们事先未知的CPU时间资源需求应该得到满足。我们使用实时Java （RTSJ）和OSGi作为实现平台。多个操作模式被添加到每个应用程序。每种模式提供不同的质量，但也有不同的资源需求。最终用户可以对应用程序进行偏好或惩罚，在其偏好的驱动下，平台通过为每个应用程序选择一种模式来实现协调和理想的资源分配。如果资源太短，不太重要的应用程序首先降级，甚至停止。科学上的挑战在于测量每种模式的实际所需资源，即使是以前未知的应用，在最佳模式的自动配置和系统对需求和可用资源的变化的动态适应。特别注意保持重新配置的数量尽可能低，并通过相移处理周期性峰值负载。

Real-time systems mostly need ex ante knowledge of required resource consumptions. This is contradictory to modern usage scenarios where applications (apps) are downloaded and installed on diverse systems, and used concurrently to each other. If applications use already existing components, an exact statement of required resouces is almost impossible. Therefore, our goal is a platform that can support applications with soft real-time requirements, e.g., multimedia applications. These can be dynamically installed and used. Their ex ante unknown resource requirements on CPU time shall be fulflilled considering preferences of the end user. As implementation platform, we use real-time Java (RTSJ) and OSGi.Multiple modes of operation are added to each application. Each mode delivers different quality but also has different resource requirements. Driven by the preferences of the end user, who can favour or penalise applications, the platform takes care of a coordinated and ideal resource distribution by selecting a mode for each application. If resources are too short, less important application are degraded first, or even stopped.The scientific challenges lie in the measurment of the actually needed resources in each mode, even for previously unknown applications, in the automatic configuration of optimal modes and the dynamic adaptation of the system on changes of requirements and of available resources. Special attention is given to keeping the number of reconfigurations as low as possible and to handling periodic peak loads by phase shifts.