[c-nsp] QOS for VoIP testing

Andre Beck cisco-nsp at ibh.net
Wed Dec 29 16:25:59 EST 2004


Hi,

On Tue, Dec 28, 2004 at 08:42:29AM -0600, Eric Helm wrote:
> What is everyone using to simulate, test, and measure QOS for VoIP calls 
> and the QOS configurations through the network?
> I have seen applications like Ixia's IxChariot 
> (http://www.ixiacom.com/products/performance_applications/pa_display.php?skey=pa_ixchariot)... 
> however I'm curious to see what is recommended and works well for others.

Probably depends very much on size and ressources of the network beeing
built. In a lab setup some weeks ago, I came up with acceptable results
by simply combining existing and free tools:

* Use ip accounting for precedence to trace how packets reach the
  routers in the correct precedence classes and leave them this way
  as well. Helped me to track a 3550 that was nulling TOS precedence
  in IP packets though only beeing used in L2 switch mode. Have a
  close look on the "mls qos" stuff here, the way it really works is
  not actually intuitive.

* Use the apropriate show commands to track the service policy and
  verify that it actually maps the traffic to the correct classes

* NBAR traffic classification might also help, if only as a debug tool

* Build an end to end VoIP setup and test that it actually works (can
  place calls) and traffic is dropping into the correct classes

* Finally use hping3 to flood the infrastructure end to end with
  data flows (UDP flood seems best) while at the same time having
  voice calls active. Monitor that using abovementioned methods
  and at the phones themselves. I've just established calls between
  7960s and used the stats display (double-"click" on the "?" aka
  Help key), looking out for RTT Jitter and especially dropped
  packets. Have a nice MoH for long-term (though unidirectional)
  testing (wildstyle and some other stations from GTAVC are great
  if you can grok XOR 0x22).

I know that there are networks where such ad-hoc homebrew methods are
not sufficient or impossible to use (the flood part is essentially a
self-DoS on anything but voice [and even on that if the config to be
tested is not complete]), but it's something free to start with.

-- 
                  The _S_anta _C_laus _O_peration
  or "how to turn a complete illusion into a neverending money source"

-> Andre Beck    +++ ABP-RIPE +++    IBH Prof. Dr. Horn GmbH, Dresden <-


More information about the cisco-nsp mailing list