[rrd-users] Holt-Winters Forecasting

John P. Rouillard rouilj at cs.umb.edu
Wed Jun 9 01:08:53 CEST 2010

In message <AANLkTiny3a_zeu9R7WY-aBQ2TSktJbhmmESXg3l1An04 at mail.gmail.com>,
John huysing writes:
>Can the Holt-Winter Forecasting  handle the fluctuating traffic flow
>that occurs over a seven day period.  The traffic pattern on the
>weekend is quite different that during the week. But the pattern from
>week to week is the same

When I was testing this I did get it to work on a 7 and larger
periods. The problem is it's kind of hit or miss on setting alpha and

What I did to test was create a perl script to generate cyclic data on
a 7 day period and then added some aberrant data on a random
basis. Then I played with the alpha and beta parms to get it to
recognize the anomolies. This was 6 years ago at another employer so I
don't have the tools anymore.

The data generating script reported how may anomolies it inserted, so
I could dump the rrdfile and count one of the parameters (failures
maybe) to see how many anomolies it detected. That means I could run
it automatically and not have to manually look at graphs to get the
best parameters.

>Also are there any more example of using   "Aberrant Behavior
>Detection with Holt-Winters Forecasting"

Not that I know of. If you come across any I would love to see them.

				-- rouilj
John Rouillard
My employers don't acknowledge my existence much less my opinions.

More information about the rrd-users mailing list