[dns-operations] Stateless-TCP performance testing tool [Was: DNS Performance Test Over TCP]

[Rick Jones]

>>     * The same client, with always the same source IP address is
>>       hammering the server.
>
> => the only drawback with this constraint (from the "standard network
> application" point) is that it is possible (i.e., too easy) to run out
> of ports. For instance you must close TCP connections as soon as possible
> on the client side.

Even closing TCP connections can still leave one short of ports.  The 
reason would be the same as when trying to do HTTP 1.0 benchmarking in 
the 1990s - attempted TIME_WAIT reuse.

Without doing something to short-circuit TIME_WAIT, once the TCP 
connection churn rate approaches or exceeds:

SizeOf(ClientTCPPortSpace)/LengthOf(TIME_WAIT)

there will be attempts to re-use a TCP connection four-tuple of 
local/remote IP, local/remote port while a previous connection by that 
same name remains in TIME_WAIT.  The IPs remain fixed, the one port is 
also fixed (53) and so the only thing left varying is the local port number.

The fix is to either use more clients and so get more IPs, or add more 
local IP addresses to the client(s) one is using.

rick jones