Posted On: Jul 25, 2019

Amazon S3 announced support for percentiles on Amazon CloudWatch Metrics. This feature allows customers to visualize and alarm on p90, p95, p99, p99.9 or any other percentile (including p100) of an S3 request metric. This provides customers with more granularity about their request patterns on S3 and helps them observe and diagnose anomalies in request patterns on S3. 

S3 request metrics already support statistics such as average, minimum or maximum, which are useful for many metrics. Percentiles are particularly useful when applied to metrics that exhibit large variances. They also help customers understand the distribution of a metric, and can be critical to understand outliers or unusual metric behaviors. Percentiles help customers observe and diagnose anomalies in request patterns on S3, avoid false alarms and save costs spent in monitoring and tracking requests. This feature is included with Amazon S3 CloudWatch Metrics for no additional charge. 

To get started, customers can turn on request metrics from the S3 console by navigating to the management tab of their S3 bucket. Once the metrics are turned on, customers can view and set up alarms on any percentile of a metric in their CloudWatch console by selecting a percentile from the drop-down list or entering a percentile of their choice, in 0.1% increments. Customers who already have request metrics turned on can skip the first step and directly go to the CloudWatch console to view percentiles of their metrics. These metrics can also be accessed using the SDK or API. 

Percentile support for S3 request metrics is available in Amazon Web Services China (Beijing) region operated by Sinnet and Amazon Web Services China (Ningxia) region operated by NWCD. 

For information about CloudWatch pricing, see Amazon CloudWatch Pricing.  

To learn more about Amazon S3 CloudWatch Metrics, visit Monitoring Metrics with Amazon CloudWatch in the Amazon S3 Developer Guide.