The transition from network operating systems to distributed global operating systems: Do we stretch too much the real-time requirements?

Abstract

The main drawback of distributed architecture is the lack of common knowledge by all processing elements, due to incompleteness of signal observability. This creates an uncertainty gap between the system-wide knowledge of every processing element at any given time, and the control functions that are required from the system at that particular time. A new approach, in hardware but mainly in software, is proposed to reduce this uncertainty gap. A reduced complexity method in the multi-processor distributed environment is presented. The mechanistic functions of the operating system are delegated from the processing sub-system (still in charge on the strategic control decisions) to the communication coordinator sub-system. A new software strategy that emphasizes efficient operation and fast decisions is presented. It's key point is task allocation strategy that overcomes possible incorrect decisions. These decisions can be reached based on partial knowledge at the time the decisions were made.