FROST is a distributed heterogeneous calculation platform first described in ``FROST - A Distributed Heterogeneous Calculation Platform''''. It provides an API that allows a user to create assignments which are distributed and calculated on a number of computers.

In this project we have chosen to improve the dynamic load balancing scheme of the FROST system by implementing process migration. We do this in order to improve the performance of the system as processes can be moved from nodes that are heavily loaded onto nodes that are less loaded during execution.

We have analyzed the problem of process migration and have decided on a solution using checkpoints as a means of performing process migration. The chosen solution has been designed and parts of the design has been implemented.

During the design phase we have been very aware that the performance aspects of FROST should not become worse after the introduction of process migration as it would remove the reason for using the feature.

We have tested the system with regard to overhead, performance and the behavior of the migration policies. The tests show that the overhead with regard to the process migration features is fairly small compared to what they offer. In addition the performance tests and the policy tests showed that the system performs well after the introduction of process migration. We do conclude, however, that in order to obtain even better performance it should be considered to introduce migration points as a tool as well.