In this project a virtual machine, gbvm, with a unique train algorithm garbage collector is implemented for the programming language gbeta. The main focus of this thesis is memory management using the train algorithm and experiments with the implementation. Firstly, this thesis introduces relevant theories and work including an introduction to gbeta, a overview of virtual machine architectures, and descriptions of garbage collection concepts, properties, and algorithms. Secondly, the design and implementation details of gbvm are presented Thirdly, experiments investigating the performance of both parameter but also algorithm changes of the train algorithm are conducted and discussed. The experimental framework implemented for this thesis allows for further experiments with gbvm. Finally, it is concluded that it is difficult to find a fixed general setting, which is both time and space efficient with all the tested programs.