[sr-dev] tcp problem

Andrei Pelinescu-Onciul andrei at iptel.org
Fri Nov 6 08:52:18 CET 2009

On Nov 02, 2009 at 19:34, Klaus Darilion <klaus.mailinglists at pernau.at> wrote:
> Andrei Pelinescu-Onciul schrieb:
> >On Nov 02, 2009 at 08:19, Klaus Darilion <klaus.mailinglists at pernau.at> 
> >wrote:
> >>
> >>Andrei Pelinescu-Onciul schrieb:
> ...
> >>>I can add the force_send_socket() sets src. for tcp new connection
> >>>feature quite easily, the question is whether or not to do it on sr_3.0.
> >>IMO choosing the src-IP dynamically is an important feature. But 
> >>probably others have to decide if it can make it into 3.0.
> >>
> >>>One difference from k would be that in k (and older ser) a connection
> >>>from IP1->X is equivalent to IP2->x (only the destination is used, if
> >>>for example you do force_send_socket(IP2); forward...(X) and a connection
> >>>from IP1 -> X already exists, this connection will be used and no new
> >>>connection will be created) while in sr the connections will be
> >>>different (both the source and the destinations are used => in the above
> >>>example a new IP2 -> X connection will be opened even if IP1 -> X
> >>>already exists which IMHO is the correct thing to do).
> >>So, it is not implemented yet, but if you would implement it, then both 
> >>(source+dst) will be used as connection identifier?
> >
> >Yes and it is already implemented (tcp_send takes both a from and a to),
> >but it's not activated.
> btw: can this be extended to take a string (e.g. domain) too - to differ 
> TLS connections which point to the same IP:port but are targeting 
> different domains (e.g. multi-domain TLS hosting using server_name 
> extension) ?

Theoretically yes. I'm not sure about the performance impact for no-TLS.
Maybe if we could get an integer id from the string (both less memory
and faster)...

Anyway let's have async TLS first and then will see :-)


More information about the sr-dev mailing list