A client of our uses a hosted phone solution for his 50 agent call centre. They use a hosted dialer who make the out going calls, once the connection is made the call is passed to my clients user where they pickup the connection on a 3CX phone interface.

Pretty standard stuff.

The hosting company are claiming that they are seeing 100-200ms latency for all the phones.

On testing a TCP round trip [tcping.exe] I'm getting sub 10ms times. They block ICMP. The hosting provider are saying that this isn't a valid test as the VOIP data is UDP.

I understand that UDP could be being shaped differently or routed differently but it seems pretty unlikely. The hosting IT team are very much 'we have hundreds of customers and their latency is ok, this is your problem' which I can understand but it doesn't help me much. Given that I can't control a UDP listener on their side I don't know how to demonstrate how and where the UDP lag might be coming in.

Equipment on site is a pair of Cisco 2960's connecting to a Draytek Vigor. The Draytek connects to the ISP provided/managed Cisco p887.

Bandwidth use sits at roughly 20% throughout the calling period. All the ICMP tests to external systems show good round trips, generally not beyond 5ms-30ms depending where I'm going to.

Thoughts on how to move this forward?

  • Are there actual call quality problems? Could their monitoring tools just be wrong? – Grant Oct 31 '14 at 12:35
  • The client is experiencing occasional dropouts, and more often than that the web interface the agents use will freeze or pause mid call which messes with how their stats tracking works. Hosting provider have said 'this is typical of high latency'. – Patrick Oct 31 '14 at 14:43
  • 1
    A web interface would be using HTTP or HTTPS over TCP so that throws any talk about UDP out the window. Other than latency, have you looked at packet loss as being the culprit? – joeqwerty Oct 31 '14 at 14:48
  • @joeqwerty yes, there is a web interface on port 80, but the VOIP data is UDP on a separate range of ports (5060 & 10000-20000). Packet loss doesn't appear to be an issue, certainly its not been brought up by the hosting provider. From my side 1000 TCP pings showed 100% response. – Patrick Oct 31 '14 at 15:45
  • OK. My point being that if the web interface is affected then the problem is not likely isolated to UDP traffic. – joeqwerty Oct 31 '14 at 15:46

Although the issue mentioned here stopped happening for no obvious reason the answer to my original question - how to effectively demonstrate network quality and roundtrip for SIP UDP traffic - is to use a SIP OPTIONS as a ping.

Brief discussion here: http://www.packetizer.com/in/q122.html

Powershell test/demonstration here : http://www.lyncexch.co.uk/sip-pinger-tool/

and I implemented this via our existing PRTG monitoring (here : http://www.paessler.com/manuals/prtg/sip_options_ping_sensor)

Unfortunatly the remote host wasn't properly configured so we received a 404 and not a 200, however we could still use that to demonstrate round trip time, although PRTG always shows this as error.


If you have a linux host on each end of the network, you can install iPerf or Thrulay to run a network test. Thrulay has the advantage in your scenerio of showing jitter during the test.

You can also watch some network stats on each end during your test or real traffic using this bash snippet:

function watch_net()
/usr/bin/watch -d -n10 --no-title 'netstat -s | grep "\ \ \ [0-9]" | sort -rn'
  • Unfortunatly I only control one end of the network so I can't do this. The remote hosting company are unwilling to make changes to their production systems for me to test against (understandably). Fortunatly the issue 'disappeared' after the hosting company performend a 'completely unrelated' network migration. Which was annoying as I was attempting to implement SIP ping to see if we could demonstrate latency/issues. – Patrick Jan 2 '15 at 9:52

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.