With the growing complexity of parallel architectures, the probability of system failures grows, too. One approach to cope with this problem is the self-healing, one of the organic computing's self-x features. Self-healing in this context means that computer clusters should detect and handle failures automatically. This paper presents a self-healing mechanism based on checkpointing, so that a cluster remains operative even if some sites or the connections between them fail. The proposed method has been implemented and tested on the Self Distributing Virtual Machine (SDVM).