I guess the main argument for doing this sooner rather than later is parallel I/O. For instance, it would be cool if we were able to run a simulation on a given number of processors, write the state of the system to disk, and then continue the simulation with a different number of processors.
I guess the main argument for doing this sooner rather than later is parallel I/O. For instance, it would be cool if we were able to run a simulation on a given number of processors, write the state of the system to disk, and then continue the simulation with a different number of processors.