I was looking at my outside temp chart, and looked at the data since I started recording this (almost 1Y ago). Then switch the display to 10Y and got a partial display, since the data before is missing (nothing fancy so far).
Then I activated all markers to see Min/Max and average. And this surprised me as the max is roughly 10°C below the actual max we experienced this year.
Is it expected that the persistence system will reduce points count with time to a single value taken a 0h00 UTC (or so it seems looking at the plots) for long term storage?
Is there a setting to have the average stored instead of the 1 point?
If you are using rrd4j, the answer is almost yes. As the data ages it gets decimated; every ten values in the db gets replaced with the average of the ten for example.
You can look at the rrd4j add-on doc to see when this decimation occurs. IIRC it’s one value every 15 minutes for year old data.
I am indeed using rrd4j, so an average does occur. However there must be some precision loss when averaging from 1Y to 10Y (or maybe this will get accurate once I have 10Y of actual data). the precision loss is ~1°C, so it is minimal.
When checking the 1Y data, it is averaged every 15’ and shows consistent values with my source database. I misundertood the max marker on the trendline. It shows the max on the plot; which is not the original max value once decimated a couple time.
Maybe you are not using the term with the specific mathematical meaning. It sounds like you mean accuracy. Precision is more closely related to number of decimal places.
Given that it’s using an average to decimate the data, I would definitely expect the values to get further from the min and max as the data ages. The average of ten values is pretty much never going to be the minimum value nor the maximum value. So the more the data ages and the more averages are taken, the farther from the min and the max the saved values will become.
It is possible to change the way rrd4j consolidates the data as it ages and MIN and MAX are among the options. But it doesn’t apply retroactively. You’d have to throw away all your existing data and start over with a new Datasource config for these Items. You’d have to decide if that is worth doing or not.
You are right, accuracy, is the proper word. In the end the difference is minimal. I will not use more effort on this. As, the original data remains in another database.
The RRD4J uses averages, and this is good enough for my use.
