From your practical experience, what is the best option for SIP service redundancy?
a) one SRV record pointing to a hostname which has two A records
b) two SRV records pointing to two different hostnames
c) high-availability tools which switch over the service IP address between two servers
I understand that c) would work with most of the clients, but many environments do not allow sharing an IP address between machines (most virtualization providers don't)
(a) and (b) are more or less equivalent. The only difference between those two approaches would concern clients that don't support SRV records. By and large, all NNI communication should support SRV records. Some user agents might not support SRV.
All DNS-based methods ((a) and (b)) incur a delay if one of the cluster members becomes unresponsive: DNS resolvers cache records, so even if the unresponsive cluster member is immediately withdrawn form DNS after it fails, some clients will still attempt to use it. High availability based on IP address will not experience this issue.
(c) generally requires either wide-area anycast (Provider-independant IP addresses and BGP) or that the high-availability cluster members be topologically close to one another (usually, on the same layer 2 network).
The two methods can be combined: multiple addresses in DNS, which each point to local-area high availability clusters.
