Ok, so then it should be checking the profiles I have added to the
holtwinter.dat file correct?  And these are profile names, not the
individual RRD filenames?

For instance, I have groups created and then profiles within the group
and each profile has multiple channels.  How should they be defined in
the holtwinter.dat file?  I currently have this:

CEN_K12/CEN_K12_Internet1
CEN_K12/CEN_K12_Web

Where CEN_K12 is the group and CEN_K12_Internet1, CEN_K12_Web are the
profiles, each of which contains 2 channels.

Didn't think I'd ever say this, but I can't wait to have an anomaly to
see how this works!

Thanks,
Robin

-----Original Message-----
From: Werner Schram [mailto:[EMAIL PROTECTED] 
Sent: Tuesday, June 12, 2007 11:14 AM
To: Brown, Robin
Cc: [email protected]
Subject: Re: [Nfsen-discuss] Holtwinter alert plugin

Hi Robin,

Brown, Robin wrote:
> I installed the plugin.  I used the --defaults directive to the
> rrd_hwreapply command to introduce the HW stuff into an rrd.
>
> Install seems successful, but I have no graph in the alerts tab for
the
> HW alert created as per the Install document.  
The graph in the alerts tab is not created by the plugin, but by nfsen 
itself and is meant for alerts conditioned on total flow summary. So it 
is not related to plugin based alerts. However, I might be able to 
addapt the plugin to replace the graphs with a custom graph. I will try 
this and let you know if I succeeded.

> I have the following in
> messages:
>
> Jun  6 11:48:00 munchies nfsen[24391]: Loading plugin 'holtwinter':
> Success
> Jun  6 11:48:00 munchies nfsen[24391]: ** Important **: Plugin
> 'holtwinter' is a legacy plugin.
> Jun  6 11:48:00 munchies nfsen[24391]: Initializing plugin
'holtwinter':
> Success
> Jun  6 11:48:00 munchies nfsen[24391]: plugin 'holtwinter': Profile
> plugin: 0, Alert condition plugin: 1, Alert action plugin: 0
> Jun  6 11:48:00 munchies nfsen[24391]: Plugin: 'holtwinter' has
> configured  profiles, but no 'run' function!
> Jun  6 11:48:00 munchies nfsen[24391]: Plugins for Alert conditions:
> holtwinter
>
> Not sure if the no 'run' function is really a problem?  
No. For plugins in the plugin tab the run function is required, but it 
is not used for alert-plugins. The plugin initialization code does not 
distinct ordinary plugins from alert plugins (yet) during this check, so

it gives this warning.

> Should the
> alerts infos graph be populated ongoing or only after an alert has
been
> triggered?
>
> Any recommendation to use other values than --defaults when running
> rrd_hwreapply?
>   
Well, at least I can tell you what I know about the parameters:

Keep in mind that the nfsen rrd data has a 5 minute time window. The 
Holt-Winters algorithm makes a prediction based on historical data using

seasonal patterns. I don't know if you know Gabor Kiss's slides, but if 
you don't, they might give a good introduction into the algorithm: 
http://bakacsin.ki.iif.hu/~kissg/project/nfsen-hw/JRA2-meeting-at-Espoo_
slides.pdf

--season denotes the length of the seasonal pattern. The default is 288,

which corresponds to 288*5=1440 minutes=1day. So the script assumes that

your nfsen data has a recurring pattern which spans one day. I think 
that this will be right in most situations.
--rralen denotes the amount of Holt-Winters data that is stored and has 
a default of 4032, which corresponds to 4032*5=20160 minutes=2 weeks of 
data. So this means the algorithm can look back 2 weeks at max.

according to
http://www-iepm.slac.stanford.edu/monitoring/forecast/hw.html :
--alpha is basically the short term parameter. A large value of --alpha 
will give a large weight to measurements very near in the past, while a 
small value of --alpha will give more weight to measurements further in 
the past.
--gamma is the trend parameter. A big value of --gamma will give more 
weight to the difference of the last smoothed observations; while a 
little value of --gamma will use information further in the past.
--beta is the seasonal parameter. A big value of --beta will give a big 
weight to the present relation between the observation and the smoothed 
observation, and little values of --beta will give more weight to past 
weeks relation between the observation and the smoothed observation.

The --deltapos and --deltaneg (both have a default value of 2) define 
the scaling factor of the upper and lower confidence band of the 
prediction. If a value is outside of the confidence band of the 
prediction it is considered a violation. So using larger values for 
these parameters, reduces the possibility of false positives, but 
increases the possibility of false negatives.
--failure-threshold (default 6) and --window-length (default 9) define 
the rate of violations that define a failure, so the default means that 
an failure is reported when there are at least 6 violations out of 9.

these pages might give some more information:
http://rrfw.sourceforge.net/rrdman/rrd_hwreapply.pod.html
http://oss.oetiker.ch/rrdtool/doc/rrdtune.en.html
http://oss.oetiker.ch/rrdtool/doc/rrdcreate.en.html

I hope this helps.

Werner

-------------------------------------------------------------------------
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/
_______________________________________________
Nfsen-discuss mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/nfsen-discuss

Reply via email to