[smokeping-users] Re: Scaling in smokeping 0.99.8

[Fred Grayson]

On Wed, 16 Jan 2002 07:24:33 +0100 (MET), Tobias Oetiker wrote:

>
>Yesterday Fred Grayson wrote:
>
>>
>> Something, somewhere, caused the latency for several sites to go way up for what looks like one 
five
>> minute reporting period. This has caused all the graphs to be scaled to that value (4 seconds). Even
>> though this happened long enough ago for the large spike to be visible on the last 30 hours and 
older
>> plots, all the plots are forced to a 4 second scale. This spoils the smoke effect :-(
>>
>> Is there any way to easily discard such data and get back to normal?
>
>Hi Fred,
>
>well not withing smokeping. You can use the rrdtool contrib
>killspike script ... another option which I am going to try soon is
>plotting with a logarithmic scale, this could make things more
>resilient to such spikes. If you have other algorithmic ideas to
>deal with this problem, let me know ... would it help if not all
>the graphs were scaling in unision ? Should there be a thershold
>for unison scaling, which lets one graph break out of the unison
>scaling as soonn as the difference goes over a certain ratio ?
>
>cheers
>tobi

Thanks for the reply. I'll chase down that killspike tool. I like the idea for logarithmic plots. That would be 
nice in MRTG too, since large transient flows have a similar effect there too, but in MRTG the shorter 
period graphs do get back to reasonable scaling very soon.

I think the non uniform scaling is a good approach too.

Thanks,

FG



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