On Thu, Feb 26, 2009 at 5:08 PM, Iñaki Baz Castillo ibc@aliax.net wrote:
However, being out there so many phones without such support, it is practically unusable since service providers won't deploy different server solutions for each group of devices, so they stick to one size fits all and that is not DNS for now.
Devices don't implement it, so service providers don't implement it, so devices don't implement it, so... XD
We use SRV extensivly, and indeed as the main means of redundancy.
However, there are lots of issues with UA implementations of SRV, most notably when they're behind NAT and jumping between different records.
Others will not ever expire it's cache, and stick to the same one. Even 2 months after we removed an RR from SRV, we're still seeing some devices using that proxy (!!).
Asterisk is particularly bad at doing SRV, to the point we've had to give it it's own DNS record:
http://wiki.voip.co.uk/products/asterisk#asterisk.bugs
There are STILL some (mainstream) devices out there that don't support DNS. this means we're forced to give IP addresses to customers, which of course is bad for redundancy and capacity planning.
As a result, we've devoted (read: wasted) a lot of time and effort on implementing anycast for our platform rather than rely on devices supporting SRV properly. It works, but it's far from pretty and something we should never have had to have done.
I'm not actually sure which is better - devices that don't support SRV, or those that try and implement it but don't get it right :-)
On the same note, it's shocking how many devices bail out after receiving a 503 and just give up. Please, implementers: 503 does not mean "never try registering again".
Proper DNS support should be enforced somehow (who knows how?!?) before anything else. At the end, DNS drives the IP world.
IMHO RFC 3263 complexity doesn't help too much.
I don't see any real complexity in RFC 3263 for a well engineered stack. What do you see as being complex about it?
~ Theo