I created a custom graph for one of my hosts to monitor the network interface. In that graph, I enabled the percentile line to see my monthly real-time 95th percentile line. Everything has gone well in the first 5 days, after that, I witnessed a significant drop in my 95th percentile line, which was weird. I looked into the graph and found that the graph was displaying less data, the sampling interval has become larger, which led to different results of the 95th percentile calculation. So I enlarge the browser window, and not surprisingly, the graph showed more data and the 95th percentile value became normal.
So the problem is, whether the 95th percentile line is calculating the result using the full dataset is depending on the window size? That doesn't make sense at all, because reducing the sample size will lead to unreliable calculation results, and there is always a limitation of the window size.
How can I get the calculation result with the full dataset(within the selected period)?
Here are the screenshots, notice I have selected the "This month so far" period and the screenshot is part of the graph, the full graph is too long to display here. The two screenshots show a huge difference in results.
The graph in smaller window size

The graph in larger window size
So the problem is, whether the 95th percentile line is calculating the result using the full dataset is depending on the window size? That doesn't make sense at all, because reducing the sample size will lead to unreliable calculation results, and there is always a limitation of the window size.
How can I get the calculation result with the full dataset(within the selected period)?
Here are the screenshots, notice I have selected the "This month so far" period and the screenshot is part of the graph, the full graph is too long to display here. The two screenshots show a huge difference in results.
The graph in smaller window size
The graph in larger window size
Comment