In this thesis, I have investigated the self-optimization approach in order to solve a generic resource allocation challenge. The challenge is defined for a general ICT system serving two classes of jobs: low and high priority. The high priority jobs require a higher quality of service compared to the low priority jobs. In order to fulfill this difference in QoS levels, a part of the total resource capacity should be held reserved to serve only high priority jobs. Two different self-optimization methods are applied to solve the challenge and the objective of the self-optimization algorithms is to split the total resources in such a way as to minimize the overall Blocking considering the different level of QoS. The first applied method is a rule-based method and the other one is fuzzy-Q learning. I also have defined a performance quality matrix which is used to assess and compare the algorithm’s reactions in three sets of designed simulation scenarios. The first set of scenarios aims to examine the effect of the different parameter settings on the overall performance of the algorithms. The second set of scenarios simulates a partial failure in the total capacity which is subsequently repaired causing the system to return to normalcy, and observes the algorithm’s reactions in adapting to these changes. The final set of simulations changes the arrival process to batch arrival where in one group of simulations the arrival rate (?) has not changed while in the other it has decreased by the rate of the average arrival batch size.