Downsampling

Definition

Downsampling reduces the number of data points in a time series to make it more manageable, without losing the essential shape of the data. It lowers storage cost and improves query performance at the cost of some granularity.

In depth

Raw metrics at 1-second resolution are valuable for real-time alerting, but keeping that resolution indefinitely is expensive. Downsampling policies typically roll up high-resolution data to coarser resolutions over time: for example, keeping 1-second data for 24 hours, 1-minute averages for 30 days, and 1-hour averages for 13 months. Downsampling aggregates (min, max, avg, count, sum) must be chosen carefully to avoid losing important peaks. obseria.io applies intelligent downsampling transparently while preserving min/max values so that spike detection remains accurate even at coarser resolutions.

Related terms