Re: [lwip-users] Question about lwip 's disconnect timeout

[David Empson]

If you are doing connect-transfer-disconnect repeatedly, you won't have a 
problem if both ends are running LWIP, as long as the number of devices 
which need to be connected simultaneously isn't too high (which is a factor 
of available memory for managing the maximum number of TCP connections and 
data associated with them).

The original article you posted talks about the standard implementation of 
closed TCP connections, where there is a delay (typically two minutes) at 
one end of the connection, with the TCP state machine held in the TIME_WAIT 
state. As noted in the article, this is to allow cleanup of any late 
arriving or duplicate packets, without possibility of confusing them with a 
new connection.

LWIP changes the standard behaviour so that if it needs to establish a new 
TCP connection (incoming or outgoing) but has run out of data structures for 
TCP connections (TCP PCBs), it will check for any closed connections still 
in the TIME_WAIT state, and reuse one of those PCBs. This effectively cuts 
off the two minute TIME_WAIT delay for the old connections. You specify the 
number of  TCP PCBs in your lwipopts.h file. This is typically limited by 
available memory in your embedded system. It would be a rare embedded system 
that could have TCP PCBs numbering in the thousands (as well as additional 
memory to buffer transmit and receive data for thousands of active 
connections), so the TCP PCBs (maximum simultaneous connections) will run 
out well before the range of 16-bit port numbers is exceeded.

Reusing PCBs in the TIME_WAIT state opens up a small risk that late arriving 
repeated packets from the old connection will be confused with a new 
connection, but that can only happen if the other station's IP address and 
port number match, and the client's port number isn't normally reused until 
several thousand new connections have been made. That is almost certainly 
impossible to achieve within two minutes on a relatively slow embedded 
server.

It is more likely to be a problem if the client made the unusual decision to 
bind to a specific port, so it uses a fixed rather than random port number 
for its end of an outgoing connection. (The server end is almost always 
bound to a fixed port number, which is not a problem.)

----- Original Message ----- 
From: "megatron" <address@hidden>
To: <address@hidden>
Sent: Thursday, June 10, 2010 9:33 PM
Subject: Re: [lwip-users] Question about lwip 's disconnect timeout


> Thanks kieran for your reply ! but because I'm a newer in TCP/IP and 
> lwip,I
> 'm still confused.
>
> In my application there is a TCP server on Lpc2368,which is an Arm7
> controller with  32K RAM ,a remote computer is client,get data from 
> Lpc2368
> via TCP .Currnetly when connected ,it will never disconnect.But there will
> be  more and more lpc2368,the computer can't connect so many lpc2368
> simultaneously.So it should be changed to connect-disconnect.TO 
> lpc2368,the
> server,be connected and then disconnected,will it stop responsing because 
> of
> the reason described above?