UDP remains the "weapon of choice" for sip trunking applications and for many subscriber applications. However, feature-rich subscriber implementations frequently require SIP messages to be substantially larger than the (approximately) 1500 byte MTU (maximum transmission unit) supported in the public internet. UDP has a mechanism for dealing with messages greater than MTU size, but it doesn't work reliably for SIP over UDP in the public internet. When you need to go > 1500 bytes, you need to go to TCP. Although SIP registration state and SIP call state can be replicated over a pair of SBC's (Acme or Sonus), TCP state cannot be maintained. There is way to much going on at the driver level for that to be practical. When switchover occurs, the TCP "transmission control" parameters are stale. So, there will be a TCP RST, and the endpoints need to start over with the whole SYN / SYN,ACK "three-way handshake" drill. Although the subscriber device will in all likelihood need to restart the TCP session, the worst-case disruption will be only as long as the re-registration interval. And even then, only outbound calls (toward the subscriber device) will be affected. But even in that unlucky instance, voice mail should save the day. The Acme "TCP session leak" was fixed several years ago At least three. Any software having that leak has been End of Support for long time. If anybody would like further discussion, feel free to contact me directly. John S. Robinson jsr (at) communichanic dot com Communichanic Consultants, Inc. On 7/17/2017 08:32, Pete Eisengrein wrote:
Perfect example. With an Acme SBC as a redundant pair, if (when) the primary fails and switches to the standby, all UDP immediately goes to the standby and is generally unnoticed by the end users. However, if the SIP is over TCP in that scenario, the switch still happens but the TCP session must re-establish itself to the secondary SBC and therefore is an outage for those users until they re-register. I suppose it is possible to share TCP session info, but I am not aware of any SBC's that do this any differently (though would love to hear from the group if one exists).
Somewhat related, but not really: It's since been patched but Acme (Oracle) had a bug at one point where it was not releasing TCP sessions after they were gone and you'd end up using all available resources (TCP ports); if that happened you'd begin blocking new sessions and the only workaround was a forced switchover which, of course, then meant a forced outage for those users.
-Pete
On Mon, Jul 17, 2017 at 9:01 AM, Colton Conor <colton.conor at gmail.com <mailto:colton.conor at gmail.com>> wrote:
Thanks for the replies. I am mainly talking about using TCP for SIP signaling for access/customer side of the network only. I think trunk connections to carriers will stay UDP only for a long time.
Overwhelming it seems like using TCP for signaling doesn't seem to be a bad thing, and preferred by many. Peter, I have a question about what you mean by "But the biggest reason I prefer UDP is for failover/redundancy. In the event of a system failure/failover, UDP will be in-disrupted but TCP will."
Lets assume we are using Broadsoft with an Acme Packet SBCs, and have redundancy having one on the West coast and one on the east cost.
Using TCP for signaling, how would this be different than using UDP in a fail over secenario? Assume the client is closer to the West cost node, and the West coast node rebooted or shut down due to power failure.
Using UDP for RTP makes perfect sense. Sorry for asking the stupid question about RTP.
On Sun, Jul 16, 2017 at 9:59 PM, Peter E <peeip989 at gmail.com <mailto:peeip989 at gmail.com>> wrote:
The SIP protocol already has some built in reliability techniques built in (timers, retransmission) for which TCP is usually used. Yes, TCP is a bit more resource intensive due to the TCP overhead. But the biggest reason I prefer UDP is for failover/redundancy. In the event of a system failure/failover, UDP will be in-disrupted but TCP will. Your TLS argument is valid, however.
RTP will always be UDP. Think about it... TCP will retransmit when packers are lost, but in real time communication there is no need to retransmit. While packet loss is problematic, a retransmission of lost packets would be unexpected and cause further quality issues.
On Jul 16, 2017, at 22:28, Colton Conor <colton.conor at gmail.com <mailto:colton.conor at gmail.com>> wrote:
I know UDP seems to be the gold standard for SIP, and is in use by most service providers that are offering hosted voice today. My question is why not use TCP instead of UDP for SIP signaling?
Overall with small business clients we run into firewalls with SIP ALGs, short UDP session time out limits, and all sorts of connectivity issues with UDP. Some small business routers and modems have built in SIP ALGs that can't be disabled at all. The second we switch to TCP for signaling most of the issues go away for our hosted voice customers. Overall TCP just always seems to work, and UPD depends on the situation of the network. TCP is better for battery consumption on mobile sip applications as well.
With more providers switching to encryption using TLS which uses TCP, is there any need for us UDP for signaling anymore? Assuming most IP phones from Polycom, Yealink, and Cisco support TCP why not use it? Is it more resouce intensive on the SBCs?
What about on the media side? Does the RTP use UDP or TCP? If it uses UDP can TCP be used? What about for encryption like SRTP? Is SRTP TCP or UDP?
_______________________________________________ VoiceOps mailing list VoiceOps at voiceops.org <mailto:VoiceOps at voiceops.org> https://puck.nether.net/mailman/listinfo/voiceops <https://puck.nether.net/mailman/listinfo/voiceops>
_______________________________________________ VoiceOps mailing list VoiceOps at voiceops.org https://puck.nether.net/mailman/listinfo/voiceops