| Abstract [eng] |
The aim of this master’s work is to generate a computational GRID clusters workload scenario tree, which would enable to solve the problem of stochastic optimization, and to optimize the working of clusters. GRID cluster scenario generation consists of the data recovery method, the cluster workload data simulation and data clustering. Data recovery was needed, since the grid cluster workload observations were incomplete. In this work two data reconstruction methods were analyzed. It was found that using different amounts of data, Maximum Likelihood Expectation-Maximization method is more efficient for the imputation of data. Time series GARCH model was used for simulation of cluster workload data paths. These data paths were clustered using a hierarchical clustering method and in this way GRID cluster workload multiple stage scenario tree was generated. |
