Using replication and checkpointing for reliable task management in computational Grids