make two checks: the system will reuse subproblems where applicable and if you run them in parallel, it should scale well