[cyberax]

My personal example: VoIP phones stopping after the Asterisk server was up for 3 days.

Reason: the server had IPv6 turned on, and it steadily accumulated privacy IPv6 addresses. These addresses were all sent in a packet describing the supported media endpoints, using UDP.

And yep, eventually it overflowed the MTU and the phones couldn't handle the fragment reassembly.

[gerdesj]

Which distro was that? ... asking for a friend ...

[cyberax]

Just regular Ubuntu. It was around 2009 or so.

[gerdesj]

Thanks. I've never run Asterisk on Ubuntu. FreePBX is CentOS based which is mostly what I've run Asterisk on. I only started to worry about Ubuntu around 2012.

That mad IPv6 address thing must have stuffed up more than just a VoIP negotiation packet. DNS switches from UDP to TCP when responses get too large.

[cyberax]

DNS is affected, but differently. It's mostly DNSSEC signatures that cause trouble nowadays.

SIP is special because the signalling and media protocols are separate. So when a call is being established, the parties exchange their media endpoint locations. This necessarily means that the server has to list its IPs (or DNS names) so that the client can choose the best one. And as a quirk of SIP, it sends the entire set for each of its supported codecs.

[gerdesj]

"SIP is special".

Yes, I know, just like ftp 8) I do hope that whomever invented putting the control channel in a separate stream from the data is mildly discomforted. Mind you, all that stuff was invented a very long time ago, when trousers were a major trip hazard.

My go to fix is "symmetric RTP", which seems to have become a default over the last decade or two.