Posted On: Jul 28, 2021
Amazon CloudWatch announces support for the trimmed mean statistic on CloudWatch Metrics. With trimmed mean statistics, customers gain visibility on the average performance of a metric without the noise of outliers. Trends in average performance represented by the trimmed mean can be visualized on CloudWatch Dashboards or used to set thresholds in alarms for proactive alerting.
CloudWatch Metrics already support statistics such as percentiles, which allow customers to monitor the worst value for a selected range, for example 95% or 99%, of the data. Trimmed mean statistics allow customers to monitor the average value for the selected percentage of data, giving a more accurate view of the users’ experience than a single percentile. This makes it particularly useful to monitor performance indicators such as latency.
For example, monitoring the tm99 would allow you to measure the average customer experience using 99% of the data and discarding only the highest 1% values, which are often skewed by outliers. Using trimmed mean statistics to monitor applications enables customers to gain visibility on a more representative sample of performance that is sensitive to variations that cannot be detected by monitoring percentiles. It thus helps address worsening tendencies early and make performance improvements visible.
With this launch, Amazon CloudWatch also adds support for percentile rank, which allows to determine the percentage of data measurements that meet a given threshold, and four other statistics: winsorized mean, interquartile mean, trimmed count, and trimmed sum.
The trimmed mean statistic is available in Amazon Web Services China (Beijing) Region, operated by Sinnet, and in Amazon Web Services China (Ningxia) Region, operated by NWCD. Standard CloudWatch custom metrics and alarm pricing applies – see pricing page for details. To get started, select a metric from the CloudWatch console and click on View graphed metrics. Under the graph, click on the Statistic dropdown and select tm99 or type tm followed by the percentage of data to include in your range, for example tm95 or tm99.999.
To learn more on those new statistics, please refer to our documentation.