RTPSniff

RTPSniff is a tool to sniff RTP traffic and show stats about it.

(I'm considering renaming rtpsniff to rtpstat -- like vmstat -- since it prints a status update every N seconds.)

TODO

• Add example zabbix template.

• Add logging to /var/run instead of syslog. That helps zabbix. (Don't forget to include options to set the user and the file permissions.)

• Also log the unixtime and interval. If needed, the zabbix reader can then divide by interval to get packets/second. And zabbix can check if the unixtime is recent enough to consider this value.

• Allow output of either all streams or only those "with issues".

• Change the interface:

  • rtpsniff 10 should show the summary every 10 seconds, like vmstat.
  • -v should make things verbose.
  • -i for the interface (take eth0 by default).
  • -B for buffer size? or -P for packets per second?
  • -b for bpf filter? or -p for pcap-style filter? and default to udp and not 53? add example for filtering even ports only? udp and not port 53 and (udp[1] & 1 = 0) and (udp[3] & 1 = 0)
  • Take into account that we may want to dump the contents.

• Decide the best form of output. Allow sequence reordering as long as it's within reasonable limits.

• Output requirements are:

  • Fix that all packets are counted, not just the seen ones, so a percentage more accurately reflects loss.

  • Always show the total packets next to the percentage. Because only a percentage is misleading when there are only a few streams.

  • A total of the RTP loss/badness; preferably a percentage.

  • List of streams with issues.

  • Current output looks like this:

        RTP: 21x.17x.21x.18x:5014 > 19x.3x.11x.10x:17886, \
          ssrc: 4022267390, packets: 43, seq: 47, missed: 5, \
          misssize: 5, late: 0, jump: 0
        RTP: 19x.3x.11x.10x:14136 > 21x.17x.21x.18x:5012, \
          ssrc: 4022267390, packets: 39, seq: 47, missed: 4, \
          misssize: 4, late: 0, jump: 0
        RTP-SUM: streams 1097, packets 214039, lost 69 (0.03%), \
          late 140 (0.07%)
    

  • For starters, the individual streams should get a loss counter.

• Move libpcap stuff out of rtpsniff.c and into sniff_pcap.c.

• Move rtp stuff out of sniff_rtp.c into cap_rtp.c.

• Document/note that timestamps are not used, only sequence numbers.

• Document/note that the streams also include RTCP.

• Features:

  • Parse RTCP from the wire and print that.

  • Create a "jitter/reorder"-buffer to store sequence numbers:

        100: [UUUUUUUUU] (91..99)
        104: [UUUUUx...] (95..103)
        102: [UUUUUx.x.] (95..103)
    

    That way we could count dupes and properly check reordering. (Store as 2bitfield? 00=unknown, 11=set, 01=skipped? init=0, increment: val=(val<<2)|3, skip: val=(val<<2)|1 oldmask=(val<<(2*seqdiff))&0x3)

Docs

• This is an adapted version of lightcount, altered to sniff RTP traffic and show which streams have packet loss. URL: https://code.osso.nl/projects/lightcount

• Simulating packet loss from the gateway:

    # 45% drop should be sufficient to get a nice robotic sound.
    iptables -I FORWARD -d SOME_IP -p udp \
      -m statistic --mode random --probability 0.45 -j DROP