Abstract

The usage development of the resources of a compute cluster varies over time. As the usage shape grows, compute resources must be procured before resource contention happens. Identifying when such a bottleneck would occur is essential for an organization that purchases and operates resources on its own. It is a domain expert who makes a procurement decision and, it becomes increasingly difficult when making a long-term procurement decision with tools such as spreadsheets. As the decision has to be made months in advance, to assist the domain expert in making a procurement decision in a transparent manner simulation-based approach is proposed. With this, a domain expert can model future resource usage development according to the business needs of the organization. In this thesis, the workload of a data-processing application is used from an industry context. The usage development can be modeled by a domain expert who estimates future demands. With the simulation, a prediction is made which identifies resource bottlenecks that would happen in the future with increasing resource usage development over time. Resources should therefore be procured to meet the demands, and with the knowledge of future usage development, a domain expert can make plans for the procurement. A case study with different scenarios is considered to evaluate the soundness of the results obtained from a long-term prediction.