[OpenSIPS-Users] LB Re: [OpenSIPS-Business] [OpenSIPS-Devel] OpenSIPS at Amoocon (former AsteriskTag)
Stefan Sayer
stefan.sayer at iptego.com
Thu May 7 16:00:53 CEST 2009
o Iñaki Baz Castillo [05/07/09 15:31]:
> 2009/5/7 Stefan Sayer <stefan.sayer at iptego.com>:
>> I must say I have not looked into the code but only at
>> http://www.opensips.org/pub/events/2009-05-04_Amoocon_Rostock/OpenSIPS_LoadBalancing.pdf
>> but I am not convinced: For load balancing that could be done
>> statelessly wrt SIP dialog you are keeping dialog state at the LB.
>
> Do you mean:
> a) Transaction statelessy
> b) Transaction statefull (but not dialog aware)
> c) Dialog aware (so also transaction stateful)
> ?
>
> With load_balancer it seems that option "c" (dialog aware) is
> required, but I don't understand if you suggest option "a" or "b". I
> understand that option "b" (transaction stateful) is required in order
> to do failover (if server 1 fails route the request to server 2).
>
at the risk of writing the obvious here, "statelessly wrt SIP dialog"
(i.e. a or b) because dialog state is much more long lived than
transaction state, and in many cases impact of a failed transaction
(because transaction statetul proxy crashed and failover one does not
have it) may be not fatal for the call due to retries, while lost dialog
state at the dialog aware proxy probably means lost call e.g. when the
next session timer reinvite comes (true or not for load_balancer with
dialog module?). so it would be the most important to not have to
maintain and sync dialog state at too many places. whether proxy is
transaction stateful or completely stateless is yet another question,
but transaction statefull is usually the way its done.
Stefan
--
Stefan Sayer
VoIP Services
stefan.sayer at iptego.com
www.iptego.com
IPTEGO GmbH
Wittenbergplatz 1
10789 Berlin
Germany
Amtsgericht Charlottenburg, HRB 101010
Geschaeftsfuehrer: Alexander Hoffmann
More information about the Users
mailing list