The distributed operating system MELODY, was developed for supporting safety critical applications in unpredictable environments. Requirements for safety and reliability have to be accommodated, as well as things like real time responsiveness. All features have to be adaptive. While developing MELODY through a number of different phases, we created a heuristic experimental, performance-relate methodology termed Incremental experimentation. Based on this novel system building approach as well as on preliminary simulation studies for non critical tasks, the focus of the paper is on different policies of file server and task scheduler integration. This is a central issue in the MELODY concept due to the specific problems of distributed file management (where copies of different "quality" are dynamically distributed over the sites). We establish three integration policies: the Periodic, Joined, and Adjusted models, and we comparatively evaluate them in distributed environments. In particular the significant results for critical task profiles are presented and discussed.
展开▼
