This explanation of this as a "second system" in the form of a "backup middlebox" feels extremely strange, as the behavior described--particularly how the "two" mechanisms are contingent: if the "primary" happens the "backup" does not (which the authors keep noting and seem to think is very surprising, due to their assertion that this is two otherwise separate middleboxes)--makes it feel like this is much more likely to be a single middlebox doing some stateful flow analysis that bails after deciding to kill a connection.