Understanding 95th Percentile Calculations for Long-Term Reports in DX NetOps PM

Products

• Network Observability

Issue/Introduction

How 95th Percentile is Calculated for 14-Day and 30-Day Reports in DX NetOps Performance Management ?

Environment

DX NetOps Performance Management 

Cause

The behavior of the 95th percentile calculation depends on the Time Range selected and the View Type (Trend vs. Table). For long-term reports, the system typically switches from raw "as-polled" data to "rollup" data (Hourly or Daily) to maintain performance, which can result in a smoother line that follows the average more closely than high-resolution raw data.

Resolution

The 95th percentile in DX NetOps is calculated using the APPROXIMATE_PERCENTILE (or PERCENTILE_CONT) function in the Vertica database. The data source used for the calculation is determined as follows:

Hourly Rollups: Uses "as-polled" data (typically 5-minute samples) from the start of the hour to the hourly timestamp to calculate the 95th percentile for that hour.
Daily Rollups: Uses "as-polled" data from the start of the day to the daily timestamp.
Table Views & Reports:

< 24 Hours: Uses "as-polled" (raw) data for the full range.

24 Hours & < 30 Days: If the user has the 'High Resolution' role right enabled, the report will attempt to use "as-polled" data.

Otherwise: The calculation uses Hourly or Daily rollup data. When calculating a 95th percentile of already-calculated 95th percentile hourly/daily values, the result naturally trends closer to the average.

Trend Views: Plot each resolution's 95th percentile value (e.g., one point per hour or one point per day) rather than a single static line across the entire period.

Exporting Granular Data: To verify the raw data points used for these calculations over a long period, use the DA OpenAPI OData query tool to export interface data to CSV/XML. This allows for external validation of the 95th percentile against the 5-minute polled samples.