Avoid cluster restart
Recent advances in Dask and Distributed may make it unnecessary to restart the cluster. This is because the memory treatment in the worker handling has been greatly improved. Avoiding the constant restarts that where necessary up to now would remove considerable overhead.