Re: [lwip-users] Recent tcp_rexmit() changes

[K.J. Mansley]

It would seem this is not a particularly new problem.  There are some
excellent notes on this (and related) issue here:

http://www.opalsoft.net/qos/TCP.htm

In particular:

http://www.opalsoft.net/qos/TCP-80.htm

talks about multiple packet losses in a single window (albeit in the
context of the fast retransmit algorithm, which is not quite what we
have now) and suggests that stacks vary in their behaviour - sounds like
lwIP is closest to Reno?

I still can't see an optimal solution - either we retransmit too many
packets if there was a single packet loss (by restarting from the lost
segment), or we don't retransmit enough if there are multiple segments
lost (by trying to restart from where it left off).  Telling the two
cases apart is going to be tricky.

Perhaps we should just implement Selective ACKs - that would certainly
help in this case! 

Kieran