I am running long simulation tasks with Dask. Each task is quite long (several seconds up to several minutes).
Unfortunately the worker nodes are spread out in different
