As IP traffic is assumed to be self-similar, my EE origins tell me to look for parameters that could measure it from stochastic process theory. On a Google search this paper sounded interesting: http://www.sparc.uni-mb.si/OPNET/PDF/IWSSIP2007Fras.pdf (...) We estimated the Hurst parameter (H) for the arrival process, and the fitted distributions for the measured data (packet size and inter-arrival processes). Using the autocorrelation function of the process, we determined long-range or short-range dependence. distribution and its parameters. The Hurst parameter was estimated using three graphical methods (variance, R/S, and periodogram methods). Distribution and its parameters were estimated using fitting tools. (...) Doing it in RRD-time seems like a challenge, though. It might be easier to plot fractals from the data source if your target audience is made of humans, because they will spot patterns real fast with much less number crunching. Rubens On Tue, Apr 21, 2009 at 9:12 PM, Crist Clark <Crist.Clark@globalstar.com> wrote:
Maybe a slightly off topic math-geek kind of question to take time out from the ARIN/death-of-IPv4/IPv6-evangalist thread of the week.
Has anyone found any value in examining network utilization numbers with Fourier analyses? After staring at pretty MRTG graphs for a bit too long today, I'm wondering if there are some interesting periodic characteristics in the data that could be easily teased out beyond, "Well, the diurnal fluctuations are obvious, but looks like we may have some hourly traffic spikes in there too. And maybe some of those are bigger every fourth hour."
A quick Google search turned up nothing at all. With many EE-types who find their way into network operations and wannabe-EEs already there, I found that maybe a little surprising. I know the EEs love Fourier transforms.