[OpenSIPS-Users] TCP Performance profiling

[Bogdan-Andrei Iancu]

HI Pete,

Sorry for the late reply on this - if this is still a hot topic, as a 
POC, try getting the rtpengine out of the picture and see how the 
numbers look for replies.

Regards,

Bogdan-Andrei Iancu

OpenSIPS Founder and Developer
   https://www.opensips-solutions.com
   https://www.siphub.com

On 15.04.2025 12:17, Pete Kelly wrote:
> I’ll give script trace a try…
>
> loadmodule "proto_tcp.so"
> modparam("proto_tcp", "tcp_parallel_handling", 1)
> modparam("proto_tcp", "tcp_async", 1)
>
> And
>
> tcp_connect_timeout=100
> tcp_connection_lifetime = 3600
> tcp_max_connections=512
> tcp_max_msg_time=2
> tcp_parallel_read_on_workers=yes
> tcp_socket_backlog=10
> tcp_keepcount=9
> tcp_keepidle=120
> tcp_keepinterval=60
>
> Although I think it must come from the TCP level as I have been doing 
> some profiling, and I see things like 180 and 200 taking 10+ seconds, 
> when there is nothing specified in the reply routes.
>
> It’s calling rtpengine_offer on the INVITE so I wonder if that is 
> causing backlogs - although the profiling never seems to raise any 
> issues on INVITE/BYE - it’s only on the replies. Maybe related to 
> transaction matching or something?
>
> Pete
>
> On 14 Apr 2025 at 10:03:38, Bogdan-Andrei Iancu <bogdan at opensips.org> 
> wrote:
>> Hi Pete,
>>
>> Try the script_trace() function, to see if the delay comes from the 
>> script level. If not, it means it comes from the TCP layer (maybe 
>> related to the dispatching of TCP cons between workers)
>> https://www.opensips.org/Documentation/Script-CoreFunctions-3-5#script_trace
>>
>> BTW, what are the TCP settings for:
>> https://www.opensips.org/Documentation/Script-CoreParameters-3-5#tcp_parallel_read_on_workers
>> https://opensips.org/html/docs/modules/3.5.x/proto_tcp.html#idp5528016
>>
>> Regards,
>> Bogdan-Andrei Iancu
>>
>> OpenSIPS Founder and Developer
>>    https://www.opensips-solutions.com
>>    https://www.siphub.com
>> On 10.04.2025 23:42, Pete Kelly wrote:
>>> Hello all
>>>
>>> I am running some OpenSIPS proxies which are talking to each other 
>>> exclusively in TCP, however when performing some (mild) load tests I 
>>> am seeing some strange behaviour from time to time, which manifests 
>>> itself as what looks like OpenSIPS taking a long time to process SIP 
>>> requests.
>>>
>>> The most obvious issue I am seeing (and confirmed by tcpdump) is 
>>> when OpenSIPS receives an invite but may take anything from 2-8 
>>> second to even send the 100 Trying!
>>>
>>> e.g.
>>>
>>> T 2025/04/10 20:28:10.257283 10.3.0.4:48665 <http://10.3.0.4:48665> 
>>> -> 10.3.0.12:5080 <http://10.3.0.12:5080> [A] #4826
>>>
>>> INVITE sip:... SIP/2.0.
>>>
>>> T 2025/04/10 20:28:12.644355 10.3.0.12:5080 <http://10.3.0.12:5080> 
>>> -> 10.3.0.4:48665 <http://10.3.0.4:48665> [AP] #4955
>>>
>>> SIP/2.0 100 Giving it a try.
>>>
>>>
>>> Is there an obvious route for me to debug this behaviour to try and 
>>> see where the bottleneck may lie ? Can I see somewhere if all of the 
>>> tcp workers are in use for example and it is waiting for a worker?
>>>
>>> Pete
>>>
>>>
>>>
>>>
>>>
>>> _______________________________________________
>>> Users mailing list
>>> Users at lists.opensips.org
>>> http://lists.opensips.org/cgi-bin/mailman/listinfo/users
>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.opensips.org/pipermail/users/attachments/20250611/8d85ffd8/attachment.html>