Welcome

The Dynamic Adaptivity in Support of Extreme Scale (DAiSES) research group welcomes you to our website. Our research spans exciting experimental and theoretical work in the areas of operating systems, high-performance systems, and automatic dynamic adaptivity. The central theme of our research is to make operating systems truly "intelligent" and automatically and dynamically adaptable. We focus our work on developing prototypes and methodologies for dynamic adaptivity in commodity operating systems, in particular for extreme-scale systems. Our research work is supported by the Department of Energy (DOE) under the FAST-OS grant and we collaborate with the University of Wisconsin at Madison and IBM-Austin's Linux Technology Center.

Explore the website to learn more about our research work and the people involved.

Cheers,

The DAiSES Research Team.

Abstract

Conventional Operating Systems, OS, manage computational resources using generalized algorithms and policies which can not be the best for all applications. In extreme scale systems, which consist of hundreds of thousands of nodes, conventional OS become a performance bottleneck. Dynamically changing resource usage patterns within and among applications, demand OS dynamic adaptation to customize the execution environment according to workload needs and system state. The main difficulty lies in understanding when and how OS algorithms, policies, and parameters should be adapted to enhance performance, and the effects of interdependencies among adaptations.

DAiSES research will provide operating systems adaptability in high-performance, extreme scale systems. Our methodology uses a priori identification of adaptation targets and effective heuristics to trigger adaptation. A combination of static analysis and runtime monitoring and adaptation provided by UW's KernInst and Kperfmon make the problem more tractable and provides the capability to evolve the OS while evolving the heuristics and the adaptations themselves.