Alex,

Yes, the issue describe: stickiness, is important.  The big problem is that pure IP-based load balancing does not work as there are two parties in a call. You may well end up with messages from A going to server 1 and messages from B going to server 2. So, you need another way to make sure that all messages in one call is handled correctly.  Here Call-Id is useful.

    A SIP (forking) proxy, the Paul (and I), interpret you is nothing more than an application level load balancer.  And as Paul points out, you have the same problems (load balancer redundancy).  In addition, for large installations, the application level load balancer will become a bottle-neck.  Then you have maintenance, stability, etc and using a well-proven high-availability load balancer seems very appealing (like LVS).  You have mainly two problems: Make sure all SER servers present the public load balancer IP address (done by using record_route_preset() and creating a bogus virtual adapter with the public IP) and stickiness in balancing algorithm

    Or did you have another approach in mind?

g-)

 

---- Original Message ----
From: Java Rockx
To: Alex Vishnev
Cc: kramarv@yahoo.com ; serusers@lists.iptel.org
Sent: Sunday, April 10, 2005 08:02 PM
Subject: Re: [Serusers] still no help - usrloc synchronization

> Alex,
>
> How does a forking proxy help when dealing with virtual IP addresses?
> Perhaps I don't quite understand what you mean by a "forking proxy"
> but it appears you'd have one "main" proxy that basically hands off
> requests to others? Right?  
>
> If so, what happens if the "main" proxy crashes? All customers
> experience a service outage.
>
> LVS is a very valid approach to achieving 99.999% uptime from a
> service perspective.
>
> Regards,
> Paul
>
>
> On Apr 10, 2005 10:49 AM, Alex Vishnev <avishnev@optonline.net> wrote:
> Hello All,
>
> I wanted to join this thread as it seems as a very interesting
> discussion. I have been monitoring for some time now. I am not quite
> sure if I this has been discussed before, but has anyone considered
> forking proxy instead of load balancer? One thing I am not very clear
> on, is what problem you are trying to resolve. Are you concerned that
> if sipua registers with sip-01/mysql-01 then LVS sending the next
> request to sip-02/mysql-02 without the second knowing what to do?
> Sorry, if this has been already discussed.      
>
> Alex
>
>
>
> _______________________________________________
> Serusers mailing list
> serusers@lists.iptel.org
> http://lists.iptel.org/mailman/listinfo/serusers