[Users] Problem with TCP window too large for TCPRCVBUF still present

[Marcin Owsiany]

On Tue, Mar 25, 2008 at 03:33:08PM +0300, Vitaliy Gusev wrote:
> Does the original kernel 028.53.3 still have issue with sending ack?

I didn't try.

> I see if VE has too small tcprcvbuf (about 30000) then server retransmit
> packets to VE. But i can't reproduce a deadlock state.

Maybe you need a crappy internet connection to be able to reproduce this
:)

The way I did that recently was:
 1) saturate the link (medium-quality 4 Mb ADSL) in both directions from
 another machine

 2) give it a couple of minutes until the transfer rates are steady

 3) start downloading a large file in a VE (I'm in UK and used a
 debian-cd mirror in Australia for that)

 4) give it a couple of minutes until the transfer rate is steady (you
 can observe the current window size in tcpdump - just wait until it
 stops changing)

 5) then stop both transfers on the other machine

 6) after some time (10-20 seconds), the testing VE will notice that
 more bandwidth has become available, and you will notice that the
 window size will start increasing, to increase the transfer rate

 7) for me, at the point the window size has reached about 9KB (it takes
 just a couple of seconds from the moment the window started
 increasing), the deadlock occured

-- 
Marcin Owsiany <marcin at owsiany.pl>              http://marcin.owsiany.pl/
GnuPG: 1024D/60F41216  FE67 DA2D 0ACA FC5E 3F75  D6F6 3A0D 8AA0 60F4 1216
 
"Every program in development at MIT expands until it can read mail."
                                                              -- Unknown