Re: SV: [lwip-users] TCP payload is doubled

[Kieran Mansley]

On Wed, 2009-10-21 at 19:23 +0200, address@hidden wrote:
> David Shmelzer wrote:
> > I'm still confused as to why the the window size is decreasing in the
> > first place in 1.3.1 and not in 1.3.0. Isn't that causing the silly
> > window syndrome? The lwIP server is processing the packets immediately
> > via the callback API so the window should not be decreasing.
> >   
> That seems like a bug in the silly window avoidance mechanism: at the 
> time your application calls tcp_recved() to release the windows space of 
> the newly received data, the stack doesn't know it hasn't sent out the 
> decreased window yet. You're right that it shouldn't shrink the window 
> in this case.

There's a very important difference here between:

1) The value for the window in the TCP header decreases as the ACK field
increases, meaning the available window goes down but the right edge of
the window is the same.

2) The value for the window in the TCP header decreases faster than the
ACK field increases, meaning the available window goes down and the
right edge of the window moves left in sequence space.

We're trying to do (1).  (2) would be illegal.  A good test is to look
at all the packets, compute the value of (ACK + WND) and make sure that
it always increases or stays the same.  This value should never get
smaller.  If it gets smaller, we have a bug.  If it always increases or
stays the same, we're fine.  The silly window avoidance code tries to
make sure that it always increases in jumps of at least 1 MSS, and to do
this it deliberately holds the right edge (ACK + WND) the same until it
can jump it by 1 MSS.

Kieran