[PLUG] UDP/ICMP packet loss question on RH

John Todd jtodd at loligo.com
Thu Jul 17 01:55:02 UTC 2003


Forgive my ignorance if this is an obvious question, but I am not a 
hardcore Linux person, and so my experiences with the Linux TCP/IP 
stack are limited.  I am, however, quite fluent with UNIX network 
protocols and this is an issue that I have not been able to 
adequately Google-search due to the number of hits I get with the 
keywords.

I am trying to track down a problem with some VoIP RTP stream jitter 
and (perhaps) loss, and the first test I did was to see if a crude 
UDP packet loss test would pass.  My results were inconclusive.

My test was to send 100 UDP packets via the 'traceroute' program, and 
see how many came back.  Yes, I know that only half the trip is UDP 
and the other is ICMP, but I figured it would be a clue that I should 
dig more deeply in that direction.

Here is what I got (IP address and hostname changed to protect the guilty):

[root at gw3 asterisk]# traceroute -f 4 -q 100 10.0.1.33
traceroute to 10.0.1.33 (10.0.1.33), 30 hops max, 38 byte packets
  4  something.foo.com (10.0.1.33)  98.391 ms  93.140 ms  92.515 ms 
94.367 ms  94.283 ms  110.650 ms *  93.908 ms  94.880 ms  101.979 ms 
94.369 ms  98.621 ms  94.046 ms *  95.153 ms  93.934 ms  94.388 ms 
94.367 ms  92.354 ms *  95.690 ms  95.359 ms  92.536 ms  96.805 ms 
94.040 ms  94.309 ms *  99.286 ms  94.270 ms  94.923 ms  96.644 ms 
99.903 ms *  93.415 ms  92.577 ms  98.413 ms  94.253 ms  93.105 ms 
93.920 ms *  98.113 ms  102.201 ms  94.287 ms  94.510 ms  98.340 ms * 
94.589 ms  93.834 ms  94.358 ms  94.460 ms  94.369 ms  92.612 ms * 
96.740 ms  95.016 ms  95.308 ms  94.643 ms  92.585 ms *  94.607 ms 
96.653 ms  95.599 ms  95.170 ms  94.210 ms  94.662 ms *  95.179 ms 
94.476 ms  92.771 ms  93.929 ms  92.455 ms *  93.142 ms  98.432 ms 
92.609 ms  95.418 ms  95.303 ms  100.843 ms *  92.146 ms  92.145 ms 
93.523 ms  93.335 ms  92.679 ms *  94.248 ms  94.181 ms  94.578 ms 
94.489 ms  92.427 ms  94.810 ms *  93.616 ms  93.377 ms  93.489 ms 
125.445 ms  100.531 ms *  92.651 ms  96.571 ms
[root at gw3 asterisk]#
[root at gw3 asterisk]# uname -a
Linux gw3 2.4.7-10 #1 Thu Sep 6 17:27:27 EDT 2001 i686 unknown
[root at gw3 asterisk]#

Note that roughly every sixth packet has been dropped.  "Very 
strange," I said, "but let's try that on some other hosts to see if 
it's consistent."  Indeed, every Linux Redhat 7.2 and Redhat 8.0 host 
I tested came back with the same symptoms, even between hosts on the 
same LAN (where, by the way, I am experiencing no problems with any 
VoIP RTP strangeness, even with the same codecs, so it's not a 
"problem".)

Now, I have been unable to answer my first question about VoIP 
badness (which I think may be software related elsewhere) but my 
investigation makes me wonder why something is dropping these 
packets, either inbound UDP or outbound ICMP.  No other OS I tested 
against showed the same symptoms.  Is this some kind of rate-limiting 
feature in Linux?  I know that IOS will not process more than a 
certain %age of packets that are ICMP outbound from the router 
itself; perhaps someone thought that was a good idea to include in 
the Linux TCP/IP stack.

(Don't harass me about the version of Redhat I'm running, or that I'm 
running Redhat.  I don't own these systems, I'm just the consultant.)

JT




More information about the PLUG mailing list