Achieving dependability in mission-critical operating systems through adaptability and large-scale functional integration

Abstract

As part of the DRAGON SLAYER project, the adaptive and reliable distributed file system MELODY has emerged for supporting hard real-time applications in unpredictable environments. In MELODY, the time criticality of tasks and their sensitivity with respect to the latest file information are explicitly used for novel, flexible task scheduling algorithms and file replication management policies, featuring dynamic file replication and relocation of file copies as well as offering file versions of varying recency. The achieved adaptability far outweighs the additional overhead in comparison to simpler models, and enhances both reliability and real-time responsiveness for file access. Both the development of the model and the experimental analysis at the various stages were done in an incremental manner. This was necessary in order to cope with the complexity of the problems resulting from conflicting requirements and dynamic trade-offs (e.g. real-time responsiveness vs. reliability). As the next incremental model extension, and as the major contribution of this paper, a series of integration policies are developed for task and resource scheduling, after redefining the role and order of task and resource scheduling: the periodic and dynamic models (and combinations thereof) are used to invoke the task scheduler for a fixed interval of time. All of these policies compare very convincingly against the classical model where the task scheduler only schedules tasks after their resources have been allocated. The results are discussed.