Suppose a client C repeatedly connects via TCP to a given port on a server S, and that each time it is C that initiates the close.
(a) How many TCP connections a second can C make here before it ties up all its available ports in TIME WAIT state? Assume client ephemeral ports are in the range 1024-5119, and that TIME WAIT lasts 60 seconds.
(b) Berkeley-derived TCP implementations typically allow a socket in TIME WAIT state to be reopened before TIME WAIT expires, if the highest sequence number used by the old incarnation of the connection is less than the ISN used by the new incarnation.
This solves the problem of old data accepted as new; however, TIME WAIT also serves the purpose of handling late final FINs. What would such an implementation have to do to address this and still achieve strict compliance with the TCP requirement that a FIN sent any time before or during a connection's TIME WAIT receive the same response?