How to detect if network is dropping UDP packets?

I'm got a video streaming application that runs fine in my office but fails miserably at the customer location. The symptom is that every couple of seconds, I stop receiving UDP packets for 2 seconds, then the stream resumes as if nothing is wrong.

I ran http://www.pingtest.net/ at the customer location and it came back excellent. No dropped packets and low latency. The only difference I noticed between our two locations is that ping google.ca times-out at their location but works in mine.

How do I test whether the network I am on blocks incoming UDP packets? Is there a way for me to isolate who is dropping the packets?

Gili

Posted 2013-04-30T15:47:43.190

Reputation: 1 559

Sounds like a firewall issue to me. Do you have any software or hardware firewalls? – Pitto – 2013-04-30T15:58:32.790

You can't ask the customer what their network configuration is set to? – Ramhound – 2013-04-30T16:05:16.797

@Ramhound, ideally not. I don't want to have to dig into a potential customer's router settings every time I want to demo my product :) – Gili – 2013-04-30T20:49:28.193

2Guys, please explain your negative votes, otherwise I can't respond. – Gili – 2013-04-30T20:50:02.650

Answers

You can try to establish a UDP connection with netcat.

On a machine A outside the consumer's network run:

nc -u -l -p 1234            # if using netcat-traditional
nc -u -l 1234               # if using netcat-openbsd (as pointed out by @JamesHaigh)

Note the -u which instructs netcat to use UDP. (And also be aware, that there are different versions of netcat, which will need the -p parameter or not; given are the variants for the two most common(?) ones, both included in Debian.)

On consumer location: nc -u [addr of machine A] 1234.

Try to send send some text, or even better use pipes to send a file between both locations and do a diff afterwards.

mpy

Posted 2013-04-30T15:47:43.190

Reputation: 20 866

Your remote command fails for me. The manpage says for -l, “It is an error to use this option in conjunction with the -p, -s, or -z options.”, so I've corrected the command to one that I've tested to work. Also, I've changed ‘ip’ to ‘addr’ because hostnames can also be used, and are in a sense an ‘address’. – James Haigh – 2013-06-18T00:07:30.657

@JamesHaigh: I agree with you on the addr vs. ip point. But now with your command, I get an error: listen needs -p arg (I tested my commands given in the answer, too ;)). There are various nc's out there, if you give some more details like nc version and/or your distro, I will add a note to my answer. – mpy – 2013-06-18T14:00:17.923

Oh right, it's a shame that they are not mutually compatible! :-( Ok, so my remote machine is Debian. The default nc command is symlink /bin/nc -> /etc/alternatives/nc -> /bin/nc.openbsd provided by Debian package netcat-openbsd. My local Ubuntu machine also has nc.openbsd by default. Neither will accept -l -p. I have also installed ncat on both machines from the nmap Ubuntu/Debian package. On the older Debian machine, ncat refuses -l -p, but ncat on Ubuntu accepts both ways. Though the Debian version must be ancient since it annoyingly doesn't have the --sctp option. :-/ – James Haigh – 2013-06-18T20:23:57.907

P.s. What is your distro, and nc variant? I notice that there is a ‘netcat-traditional’ package as well, but I haven't tried that. – James Haigh – 2013-06-18T20:30:14.303

@JamesHaigh: I indeed use netcat-traditional (v 1.10-38) as it ships with debian. Thanks for your hint, I included now both variants into the answer. – mpy – 2013-06-19T10:37:36.897

According to /etc/debian_version, my VPS is ‘6.0.2’. What version of Debian are you running? – James Haigh – 2013-06-19T14:46:34.157

@JamesHaigh: I also have wheezy, but version 6.0.1. Btw. OpenSuSE uses traditional netcat, too. – mpy – 2013-06-19T14:59:32.997

You mean squeeze? It strikes me as odd that Debian would change defaults in a point release – maybe it was a DreamHost-specific customisation. – James Haigh – 2013-06-19T15:22:30.517

@JamesHaigh: Sure, squeeze, sorry. However I checked on wheezy (7.1) installation as well. apt-get install netcat installs again netcat-traditional (v1.10-40). It sounds plausible that netcat-openbsd is a non-standard default for DreamHosts. – mpy – 2013-06-19T20:59:42.223

at server side, establish a UPD server with

iperf -s -u

at client side, check UDP connection with

iperf -u -c <IP Address of Server>

ckarrie

Posted 2013-04-30T15:47:43.190

Reputation: 131

1This is the real answer. It requires having server access on the other side. But it gives you direct packet loss feedback. And you can also use it test TCP bandwidth. – DragonFax – 2017-07-05T17:12:10.057

The netcat commands in mpy's answer are useful for diagnostic purposes, but I'm complementing that answer with another approach to your underlying problem.

It may be worth making your application fall back to SCTP, or even TCP. I actually found this question because I was looking for how to reject incoming UDP packets from users using more than their share of the downlink when it is congested, because unlike SCTP and TCP, UDP has no congestion control making it very difficult to prioritise the downlink traffic.

Both SCTP and TCP have congestion control and play nicely with QoS, but SCTP has the additional benefit over TCP that it was designed for real-time streaming applications, making it a good replacement for both TCP and UDP. In effect, SCTP is the best of both of the 2 most common transport protocols.

It can't be a bad idea to have a fallback, rather than to rely on just UDP. Even if you only fall back to TCP, then at least you can say it's working, maybe just not optimally.

James Haigh

Posted 2013-04-30T15:47:43.190

Reputation: 554