Hello Daniel,
Indeed xlog doesn't print in syslog the sip message received and wireshak got it, that means that packet was received in network interface but kamailio didn't received the packet(as xlog didn't print that packet).
Anyway i have configured the following parameters in sysctl
net.ipv4.udp_rmem_min = 131072
net.ipv4.udp_wmem_min = 131072net.core.netdev_max_backlog=1000net.core.rmem_max=67108864net.ipv4.udp_mem = 192576 256768 385152
and also set maxbuffer=134190336 in my kamailio configuration.
With these configurations I got an improvement but still below of what we need and even the CPU is between 65% and 70% sleeping.
Do you have any suggestion? the traffic average that interface receives now is 200mbs.
The server network interface is 10GB.
Any suggestion is welcome.
Best RegardsJosé Seabra
2015-09-27 17:08 GMT+01:00 Daniel-Constantin Mierla <miconda@gmail.com>:
Hello,
first, be sure you run latest version of branch 4.3, there was a fix for dialog since 4.3.2.
If a process in shown to run recvfrom() that means it waits for kernel to pass a new upd packet to it -- recvfrom() is a function from standard lib c.
Wireshark (like other sniffing tools) is capturing the network traffic before the kernel passes it to application layer. If wireshark see it, doesn't mean the application has received it from kernel, it can still be dropped by firewall or other kernel modules.
You can add a xlog() at the beginning of request_route to see what requests are received by kamailio.
Cheers,
Daniel
On 27/09/15 15:52, José Seabra wrote:
Hello Danel,Thank you for your response,
Well i don't have SELinux enabled and IPTables isn't running.
About kamailio configuration, Pike module isn't also enabled, the only thing that i have enabled to limit the number of concurrent calls is the dialog module, counting the number of active dialogs and reject next new invites only after it has reached the 10.000 simultaneous calls.
The strange behavior is that kamailio only stop processing some sip messages like Byes and 200 Ok after more or less 40 seconds after siptester start sending calls.
And i can see several cases in wireshark that kamailio receives BYES/200OK from interface B(Carriers interface) but it isn't forward to Interface A(customer interface), if i make for example 20cps i don't have this issue, all packets are handled correctly and there isn't 408 requests timeouts.
Best RegardsJosé Seabra
2015-09-27 13:52 GMT+01:00 Daniel-Constantin Mierla <miconda@gmail.com>:
The tmp variable is not intialized by default, so it can be anything therefore the out of bounds value. It used only when printing some errors, so it didn't get a proper value as no such error was printed for you case.
The backtrace shows the process in in recvfrom(), so it waits for traffic from network. That is the case when the processing of former sip message was finished and now Kamailio waits to receive another one.
If you run centos, be sure you don't have limits in selinux rules - better disable it completely and test again. Also, be sure you don't have limits in the firewall, especially contrack module and sip firewall modules in the kernel.
In kamailio.cfg, if you have pike module, be sure you don't drop good packet with this module.
Cheers,
Daniel
On 25/09/15 19:53, José Seabra wrote:
Adding more information to my last email, when kamailio starts receiving calls, after more or less 40 seconds it stops processing some sip messages.I'm using the kamailio version 4.3.2
Best Regards
-- Daniel-Constantin Mierla http://twitter.com/#!/miconda - http://www.linkedin.com/in/miconda Book: SIP Routing With Kamailio - http://www.asipto.com