[rrd-developers] Re: Instability in HWPREDICT ?

Stanley Hopcroft Stanley.Hopcroft at IPAustralia.Gov.AU
Thu Aug 29 01:56:40 MEST 2002

Dear SIr,

I am writing to thank you very much for your reply and say,

On Wed, Aug 28, 2002 at 08:29:50AM -0700, Jake Brutlag wrote:
> Stanley Hopcroft,
> My best guess is that the "lump" in DEVPREDICT is probably due to an
> order of magnitude shock (a spike or spikes) in your time series at that
> point (or near that point) in the seasonal cycle at some point in the
> past. If you have a daily cycle (the default with 5 min intervals and a
> seasonal cycle of 288), then this could have occurred in the last
> several days. The algorithm has "learned" that there is unpredictability
> at this point in the seasonal cycle and therefore has a wide confidence
> band.

Yes, I think so too.

The time series originates from processing that is occurs mainly in 
business hours; the numbers processed on Saturdays and Sundays are very 
small (and in this case, even lower than usual, last weekend because the 
process failed) compared with those during the week.

It seems that in this case, I would be better off either

.. only collecting data on weekdays (this losing the opportunity of 
detecting weekend faults), Or

.. 'tuning' alpha and beta as you recommend.

> By default the seasonal deviation tuning parameter will be set to the
> same value of alpha (the intercept adaptation parameter). With a value
> of 0.0035, it might take some time to "unlearn" the spike. You can try
> setting the seasonal adaptation parameter to a more aggressive value
> (via the tune command with the -v flag). Our production environment is
> using the value 0.28.

Thank you very much for the hint. It would/will take me a long time I 
suspect to understand HW prediction to this extent.

Would you care to suggest any readable (under graduate stats background,
weak but functional maths) explanations or references ?

> BTW, Stanislav Sinyagin [ssinyagin at yahoo.com], a contributor to this
> list, has developed a tool which allows you to experiment with different
> tuning parameters for aberrant behavior. You may want to contact him and
> play around with it.

Wow. Thank you.

> Jake
> Jake Brutlag
> Network Analyst
> TV Services -- Network Operations
> Microsoft MSN 

Yours sincerely.

Stanley Hopcroft

'...No man is an island, entire of itself; every man is a piece of the
continent, a part of the main. If a clod be washed away by the sea,
Europe is the less, as well as if a promontory were, as well as if a
manor of thy friend's or of thine own were. Any man's death diminishes
me, because I am involved in mankind; and therefore never send to know
for whom the bell tolls; it tolls for thee...'

from Meditation 17, J Donne.

Unsubscribe mailto:rrd-developers-request at list.ee.ethz.ch?subject=unsubscribe
Help        mailto:rrd-developers-request at list.ee.ethz.ch?subject=help
Archive     http://www.ee.ethz.ch/~slist/rrd-developers
WebAdmin    http://www.ee.ethz.ch/~slist/lsg2.cgi

More information about the rrd-developers mailing list