Re: [lwip-users] strange problems (small etherreal log attached)

[K.J. Mansley]

On Mon, 2004-11-01 at 07:59, Bonny Gijzen wrote:

> In this log the missing frame was detected 10 seconds later (at the
> time my client had sent a new msg).
> This is wrong because lwip would need to see an ACK and if not a
> retransmission should kick in much earlier.
> (I am not sure about the timeframe used here, but should be millisecs
> I guess)
>  
> I have checked my lwIP timers, they are OK.
> (I use a 250msec interval for tcp_tmr() which invokes both the fast
> and slow timer)

You're right that the retransmission should happen without receiving
another packet.  After a few calls to it (so probably at least 1
second), the slow timer routine should notice that a packet remains
unacked, and retransmit.

Although you say the lwIP timers are OK, it looks like this is where
things must be going wrong.  I would insert some debug statements
liberally in the slow timer function and hope that sheds some light on
your problem.  It should go through the list of segments on the
"unacked" queue of each pcb and increment a counter.  When that counter
reaches a threshold it knows it's been hanging around for too long and
it should then be retransmitted.  Check that this is all happening
correctly.

Thanks, and sorry that I don't have more time to investigate.

Kieran