Simulation-Aided Decision Support for Cooperative Resource Sharing

A number of problems, concerned with resource sharing, have to be solved when considering a cooperative resource sharing environment. One problem is the question, how many resources of a participating site should be contributed to the environment. This problem can be addressed by a simulation-aided decision support system. The idea of such a system is to create different what-if scenarios and simulate them with the aim to see, how a certain decision influences the overall system behaviour. Based on the simulation results a decision regarding resource contribution can be made. For the simulation of what-if scenarios it is necessary to have a precise model of the cooperative resource sharing environment which can be used for the simulation. In addition, it is necessary to have accurately predicted workload which can be used for simulations. This thesis is concerned with modeling of a cooperative resource sharing environment and workload prediction. A number of definitions regarding resources and resource sharing policies are given. The definition of a resource described in this work is not limited to a certain type of resource. The described resource sharing policies include capacity sharing and time sharing as well as a hybrid sharing policy. In addition, an event-driven simulation model is explained which can be used to describe the scheduling of jobs on a site. For workload prediction, a number of available workload logs and prediction approaches are analysed. A number of workload prediction models are proposed as well as a logging and prediction framework, with focus on flexibility and extensibility. Although accuracy is lower for heterogeneous environments, validation has shown the capabilities of the event-driven driven simulation model to describe scheduling of jobs with a high degree of accuracy in general. The results gained from experiments with different generic workload prediction models have shown a very limited accuracy. Accuracy of simulating what-if scenarios highly depends on the precision of the simulation model as well on the accuracy of predicted workload. A relatively precise simulation could be presented in this work. The workload prediction models presented in this work seem to be too generic for highly accurate workload prediction. Further research is therefore necessary.