Scheduling Analysis With Java Modelling Tools

Queueing and scheduling theory have both been extensively dealing with performance measures like response and residence times, uti- lization, throughput and other related metrics. I have investigated the characteristics of a relatively new measure that has recently re- ceived growing attention; fairness. I have attempted to evaluate pro- posed definitions of fairness in practice, by extending Java Modelling Tools (JMT) to support both the new metric of fairness and also new scheduling policies that would allow useful comparisons. In partic- ular, I have implemented the following policies which are ready to be released: Random, Random with priority, Shortest Job First, SJF with priority, Longest Job First, LJF with priority, Weighted Prior- ity, Shortest Remaining Processing Time. JMT is a very powerful software which has a large community of students, researchers and practitioners and therefore this project will benefit many. However, as many people, with different programming and commenting styles, have worked for the development of the hundreds of classes that con- stitute JMT, the technical complexity of the project was significant. The main challenge though, was to properly evaluate the results of my implementations. For this reason, I have conducted numerous exper- iments, compared my results with other existing software, used test classes and analytical methods.

Some interesting results from my investigation include that the fair- ness of a scheduling policy differs a lot between different fairness metrics, which is a proof of the subjectiveness of fairness. Moreover, as the utilization of a queue increases the relative fairness of the scheduling policies is not affected, but the fairness of a particular policy either increases or decreases according to the metric being used.