The design of an adaptive, intelligent operating system scheduler

This paper describes the design and implementation of an adaptive, intelligent operating system scheduler, based on machine-learning of heuristics. This scheduler is termed the adaptive scheduler controller (ASC). The question which the project seeks to answer is whether expert system technology, specifically the technique of machine-learning of heuristics developed by Waterman (1970), can be successfully used to create an adaptive scheduler. Such an adaptive scheduler could be expected to operate in an environment which is not completely determined. The procedure for machine-learning of heuristics is as follows. The ASC makes a series of decisions and then observes the behavior of the system as determined by these decisions. From this feedback the ASC deduces what the correct decisions should have been and compares the correct decisions to the ones it actually made. For each incorrect decision the ASC will modify its set of production rules, or heuristics, so that the incorrect decision will not be made in the future. The results are that the ASC was able to adapt successfully to changes in its environment and to its own behavior. It learned rules to correctly assign priorities to processes even though the behavior of these processes changed. It also was able to recover from errors which it made in assigning priorities. One may conclude that the techniques of machine-learning of heuristics can be used to create an adaptive scheduler. This technique may find wider applications to systems programming problems in which the system is required to make complex decisions in a variable environment