[rrd-users] VDEF, 95th percentile and graph width

Marius Karthaus lists at karthaus.nl
Fri Sep 5 12:36:23 CEST 2014


Hello Johan, Martin and Steve,

Thank you for your thoughts.
To make it more clear I've put the graphs and some data here: 
https://noc.lowvoice.nl/test/stats

I get the idea that the graph is made up of the datapoints aggregated 
for the needed pixel width. But for the VDEFs  it should operate on the 
full dataset according to the docs: *"VDEF*instructions work on an 
entire data set in one run.", so pre-aggregation should not be needed. 
Pre aggregation, in the case of 95th percentile, causes problems. (pre 
aggregating datapoints as average and then getting the MAX is also a 
good example of why this should not happen).

I understand that you eventually lose resolution as a tradeoff between 
storage-space and accuracy, but I do not understand why VDEFS would 
choose not to use all the available data.

The 'step=300' trick, as suggested by you all and the 'use different 
DEFs & VDEFs for the graph part and the gprint part' as suggested by 
Steve, have no influence on my results as you can see on the above page.

Any additional thoughts would be appreciated.

Regards,
Marius Karthaus





-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.oetiker.ch/pipermail/rrd-users/attachments/20140905/d0ae4882/attachment.html>


More information about the rrd-users mailing list