[stunnel-users] RFC 2487 test in smtp_server not reliable

[Hans Werner Strube]

Hans Werner Strube wrote: 
> Michal Trojnara wrote:
... 
> > I'm going to do it the hard way and add another parameter to 
> > s_poll_wait() function.
> > 
> > IMHO the timeout should be between 100ms and 500ms.  What do you think?
> 
> I have not tested the minimum required. However, it should be as short
> as possible, since for a client using TLS, the full delay will always
> occur on starting a connection. ...

I have now done some measurements using snoop on a Solaris Ultra 2 SMTP
server (100 MBit/s) with a Windows XP SP2 client (AMD Duron 1 GHz, 56 kBit/s
phone-line modem, SeaMonkey 1.04 mail sender [equivalent to Thunderbird
1.5.x]).
The delays between the end of initial connection handshake and the first
nonzero data packet for six connections were usually between 10 and 100 ms
but once reached 150 ms. Thus, 200 ms may be a reasonable value (although
the sample is too small to rule out rare failures).