[e2e] Re: crippled Internet

Christian Huitema huitema at exchange.microsoft.com
Fri Apr 27 17:21:48 PDT 2001


> Kim Claffy wrote:
> 
>   Can we get better data?
> 
> christian
> can you (or someone) be more specific?

To predict VoIP quality, we need to understand the structure of the tail
of the delay distribution. Is it Poisson like, or more heavy tailed? The
classic way to predict that is to plot the cumulative frequency
distribution on a log log graph; most probably, the tail will look like
a straight line (power law) or drop faster (Poisson); if it is a power
law, the slope of the tail gives you the exponent. Note that this is
also useful to understand mundane stuff such as the behavior of TCP
timers.

-- Christian Huitema



More information about the end2end-interest mailing list