Inaccurate timestamps on weekly/monthly graphs?

[olivluca]

Edit: funny how you find the explanation to the question below right after clicking on "post thread" , it's probably because the weekly graph has a 3 hours resolution and the monthly view a 12 hours resolution.

I have a backup job scheduled at 5:25 in the morning and it takes about 20 minutes.
In the daily graph, I see a peak in cpu usage (and network output) at 6:00, which is about right


But it I switch to the weekly view the peak is at 8:00



and in the monthly view the peak is at 14:00



It's not a problem, I'm just curious about this time shift between the daily/weekly/monthly views.

 

[UdoB]

It is using a round-robin-database --> https://oss.oetiker.ch/rrdtool/index.en.html

The main aspect is that the size of that database is (nearly) constant, well predicable and relatively small. To see current data with a higher resolution than historical data it condenses it continuously. The resolution and the accuracy is not constant but uses different scales, especially with a fixed time stepping per (hard, during installation, not changeable) defined intervals. This produces rounding errors.

For this reason "old" data is visible only in coarse steps.