We have been using a histogram grouping power of 7 internally, but we can often tolerate higher relative error for metrics reported out of these histograms.
This PR reduces the histogram grouping power to 3. This gives a 12.5% relative error and reduces the per-histogram size from 58KiB to 4KiB. Our previous relative error was bounded at 0.781%.
We have been using a histogram grouping power of 7 internally, but we can often tolerate higher relative error for metrics reported out of these histograms.
This PR reduces the histogram grouping power to 3. This gives a 12.5% relative error and reduces the per-histogram size from 58KiB to 4KiB. Our previous relative error was bounded at 0.781%.