Re: [lwip-users] Decreasing window size problem

[Kieran Mansley]

On Fri, 2007-06-08 at 11:41 +0200, Per-Henrik Lundblom wrote:
> * Kieran Mansley <address@hidden> [070608 11:13]:
> 
> > On Fri, 2007-06-08 at 10:58 +0200, Per-Henrik Lundblom wrote:
> > 
> > > Mmm, the probes are basically splitted packets from the unsent queue
> > > with a data length of 1 byte. So, the splitting must still be performed
> > > =/
> > 
> > Good point.  The reason I'm reluctant is that in the normal run of
> > things (where the window update doesn't get lost) not splitting the
> > packet saves the sender a fair bit of work without really causing
> > performance problems.  So I'm happy to go to the effort of splitting the
> > packet if we think there's been a problem (i.e. we're sending a probe)
> > but would rather avoid it if everything is OK.
> 
> I understand that, splitting the packets adds unnecessary complexity but
> I don't see another solution to my problem.

Splitting to send a probe when your problem occurs is fine.  We need to
treat "I can't send because the window is too small" the same as "I
can't send because the window is zero" and send probes in both cases.
At the moment I think we only send probes in the latter, although I've
not checked the code so could be wrong.

Kieran