[Freeswitch-users] RTP packet loss on outgoing streams

mbo mbodbg at gmx.net
Fri Apr 17 19:01:01 MSD 2015


Hello,

we need to monitor the voice quality on our server. To test this, we setup a test environment with 2 freeswitches, originating calls from freeswitch 1 and then answer them on freeswitch 2 and play an announcement. To analyze the traffic, we us tshark with the following command:   

tshark -q -f 'udp' -o rtp.heuristic_rtp:TRUE -z rtp,strems -w test-pcap

The result is, that we see packet loss around 20% on outgoing RTP streams, when we have more then 20 calls. The load average shown with the top command is always < 0.3 with a 4 core CPU and CPU load is less then 10%. 

========================= RTP Streams ========================
    Src IP addr  Port    Dest IP addr  Port       SSRC          Payload  Pkts         Lost   Max Delta(ms)  Max Jitter(ms) Mean Jitter(ms) Problems?
  192.168.1.134 27590   192.168.1.135 22576 0xCD7A189D      ITU-T G.722   800     2 (0.2%)           59.93            0.11            0.05 X
  192.168.1.135 22576   192.168.1.134 27590 0x5D4E50CD      ITU-T G.722   599   202 (25.2%)          160.24            0.16            0.05 X
  192.168.1.134 28172   192.168.1.135 29704 0xD53EC9AD      ITU-T G.722   798     2 (0.2%)           59.99            0.07            0.04 X
  192.168.1.135 29704   192.168.1.134 28172 0x5D50942D      ITU-T G.722   713    86 (10.8%)          199.98            0.08            0.04 X
  192.168.1.135 25794   192.168.1.134 16460 0x5D51E43D      ITU-T G.722   633   165 (20.7%)          180.25            0.75            0.07 X

Instead of tshark, we also tried to capture SIP & RTP with tcpdump with the following command:

tcpdump -i any -T rtp -vvvvv -s 0 -w test.pcap

If we analyze this file with wireshark (TELEPHONY->RTP->Show all Streams), it shows almost always 0% for packet loss. Only if we increase the load to 50 calls we can see a similar scenario where inbound calls still have < 1% packet loss but outgoing RTP streams have up to 20 % packet loss. We also repeated the same test with freeswitch 1.4.15 with much more powerful hardware with almost the same results.

The question is what do we see here. Does tshark show us wrong result? Can we more trust on the results captured with tcpdump or is there finally an issue in freeswitch when sending outgoing packets? Or is the method to capture on the server in general the wrong approach? 

Thanks for any comments and hints

Markus

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.freeswitch.org/pipermail/freeswitch-users/attachments/20150417/ccea19de/attachment-0001.html 


Join us at ClueCon 2016 Aug 8-12, 2016
More information about the FreeSWITCH-users mailing list