On a given node, we maintain a few million sorted sets that see a query and insertion velocity in the orders of millions of operations per second. When we are meticulously measuring operation time in the microsecond range to meet a performance target, a networked solution is out of the question.