we use a consistent hashring to try to run the same subproblems on the same nodes, ensuring they can use the in-memory cache