[rrd-users] Re: How to eliminata extrimely high values

[Alex van den Bogaerdt]

On Sat, Sep 28, 2002 at 01:29:41PM +0300, Katsifos Nikolaos wrote:

> I am having a graph but sometimes this graph recives  extrimely high values.
> This is shows my graph very strange (a flat line at the bottom and a peak
> somewhere)

First have a look at your "extremely high" values.  Perhaps you
have a configuration problem and should get rid of those values
altogether.

For instance: If you didn't set a proper MAX setting, you can expect
random high values when the device reboots or when the counters are
reset.  These values can be *very* high, such as terabytes per second.

If you suffer from this, don't try to work around the problem but
solve the problem.

> Ofcourse i can use LIMIT like this.
> 
> CDEF:V=value,0,1000,LIMIT
> 
> This will eliminate my stupid peaks
> 
> But this will not calculate my values automaticaly and i have to give a
> diffrent LIMIT every time i ploting my graph.

This is doable.

> IF (VALUE EQUAL WITH THE  5% OF THE HIGHEST VALUES )
> THEN (VALUE = AVERAGE)

This not.


What you can do is write a script that computes an appropriate
value for the LIMIT calculation.  Then use this value in the
rrdtool graph script.

You can write a 95th percentile calculator by using rrdtool fetch,
sort, wc and sed.

Use rrdtool fetch to get the values you'll be going to graph.
Then sort the values using the sort command.  Use wc to get
the number of rows (or: *know* how many rows will be returned).
Now use 'sed -n 12345p' to print the line with the 95th percentile.
(where 12345 is a suitable number).  The resulting IN and OUT
values can be used in the "CDEF:V=value,0,xxx,LIMIT" line.

However:  How do you know that the 95th percentile doesn't result
in another high value?

Alex

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