# Fixing noisy prometheus grafana dashboards with quantile\_over\_time (2024-03-24)

**URL:** <https://pluralist.net/t/fixing-noisy-prometheus-grafana-dashboards-with-quantile-over-time-2024-03-24/119>\
**Category:** breadcrumbs\
**Created:** [March 24, 2024, 4:00am UTC](https://pluralist.net/t/fixing-noisy-prometheus-grafana-dashboards-with-quantile-over-time-2024-03-24/119 "2024-03-24T04:00:00Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![delyan](https://pluralist.net/user_avatar/pluralist.net/delyan/32/245_2.png) [@delyan](https://pluralist.net/u/delyan)\
**Post date:** [March 24, 2024, 4:00am UTC](https://pluralist.net/t/fixing-noisy-prometheus-grafana-dashboards-with-quantile-over-time-2024-03-24/119/1 "2024-03-24T04:00:00Z")

</div>

Our home Grafana dashboards are wonderful - we are monitoring ping latency to hosts we care about, as well as humidity and temperature.

The ping latency graphs are difficult to parse at a glance, however. There is too much noise, which obfuscates the meaningful information we need.

Here is the basic prometheus query and the resulting noisy ping latency:

```auto
ping_gauge{destination="8.8.8.8"}

```

 ![Screenshot 2024-03-24 105657](https://pluralist.net/uploads/default/original/1X/ca2a1a599acc72af0025ec39cbc9fff91eb07398.png)

Using the idea from [Quick tip: How Prometheus can make visualizing noisy data easier | Grafana Labs](https://grafana.com/blog/2020/10/20/quick-tip-how-prometheus-can-make-visualizing-noisy-data-easier/) we are going to apply `quantile_over_time` to de-noise.

Here is the new prom query and the resulting “smooth” grafana dashboards:

```auto
quantile_over_time(0.5, ping_gauge{destination="8.8.8.8"}[1h])

```

which then results in much smoother and easier to grok-at-a-glance (ping) graphs:

 ![image](https://pluralist.net/uploads/default/original/1X/6d0847cad8f2e03a29c93d638e1a60f0db649b67.png)

## Conclusion

Applying `quantile_over_time` to fairly noisy data such as `ping_gauge` (latency time series data) we were able to remove noise, and smooth the data to a point where large-scale events become obvious. Now we can confidently ask the question: _What happen to the Internet connectivity between continental US (California & Michigan) and Bulgaria from 9:45pm until midnight Eastern time zone._