[Serdev] Concept for a SIP cluster implementation

[Jiri Kuthan]

At 08:44 AM 9/28/2005, Greger V. Teigre wrote:
>>I don't want to use SRV, not only because of the NAT problem but also because most of the SRV implementations in UACs I've seen so far are broken, if implemented at all.
>>
>>So each edge proxy will be set up with a hot-standby proxy for IP failover, so there's no need for a failover mechanism in the UAC (to be as much independent from it as possible).
>
>And back to one of my favorite topics: Avoid idle hardware.  Andreas, you talked about either linux ha or quagga and bgp as failover mechanisms.  Is there a way to combine a hot and a standby server in the same physical server (without vmware/xen)?

I am not sure that's what anyone would like -- host failures happen too,
e.g., harddisks happen to go.


>>If the first edge proxy pair (say sip1.my.domain) get's to a limit regarding ressources, you can add a second pair and propagate the new hostname (sip2.my.domain) to new customers.
>
>And of course, if you do happen to have UACs with good SRV support, you can use that :-)

That's indeed the major problem with SRV: most clients don't get it right
and resolving to different IPs using SRV has other issues with NATs.

The bottom line is that preserving server IP address to make NATs happy 
even on failure is not entirely easy without having idle hardware. 

-jiri