Re: is tcp flow control broken in -current

[uwe%stderr.spb.ru@localhost (Valery Ushakov)]

Frank Kardel <kardel%netbsd.org@localhost> wrote:

> The application runs and processes large number of queries via the tcp 
> back-end connection via the loopback interface. After a while the 
> communication slows down significantly.
> 
> The database back-end get a single character of a query string once 
> every five seconds and later on once per minute. At the same time 
> netstat -an
> 
> shows that the send-queue has more than one character available.
[...]
> I suspect something triggered the slow sending and it seems to be stuck 
> in that state even though
> 
> the window is 4096 and all packets are acked.
> 
> Any ideas?

Without actually looking at the details...  I've fixed a bug a while
ago that was kinda similar:

  http://gnats.netbsd.org/49390

I wonder if you hit something like that too as sending 1 byte at large
intervals looks like a window probe.  Except nothing in your trace
suggests you should be in that territory.  Well, you asked for *any*
ideas... :)

-uwe