Analytics
When you create a new source in HCM, you can enable Anomaly Detection and Forecasting.
Anomaly Detection
Anomaly Detection monitors your HCP system to detect and report any unusual storage and network usage readings across the system. Every new reading is compared to a distribution of historical readings using percentiles. A reading is considered an anomaly if it is highly unusual compared to historical data, such as being above the high extreme or below the low extreme.
After an anomaly is detected, the Monitor App sends you an alert that you should investigate. An anomaly could be expected if you have added an application that uses a lot of storage. Or, it could indicate that a service is not performing as expected. The alert clears only when sufficient time has passed and the percentiles of readings have shifted so that the original reading is no longer unusual.
When you enable Anomaly Detection for a source, your system automatically creates two monitors for the following conditions:
- Anomaly Storage Used: Dramatic changes in the hourly system storage, whether an increase or decrease
- Anomaly Front-End Total: Dramatic changes in the front-end network traffic
Forecasting
Storage forecasting predicts when your system's storage capacity will be reached based on historical HCP data. A Bayesian time series model incorporates weekly and monthly season patterns to predict future storage usage, with a confidence band attached to it. A new storage forecast is generated once a day.
For every source you enable Storage forecasting on, the Monitor App shows a graph of the storage usage, along with the following information:
- Capacity: The maximum amount of storage in a system
- Confidence: The degree of certainty that the actual storage used on a future date will be within the confidence band's range.
- Days Remaining: An estimation of the amount of days left until the storage capacity is reached. This estimate is based on the date that the confidence band and the capacity level intersect on the graph. Your system is expected to run out of storage after that date.
To see the forecast for a system, click the Analytics window and select the source.
The SSD Health forecast predicts when your system's storage capacity will reach a warned state for a failing drive using SSD wear metrics from HCP. When SSD Health is enabled, the Monitor App displays a trend graph of the remaining useful life (RUL) of the SSD drive predicted to reach end-of-life the fastest.
The following information is displayed on your forecast:
- SSD Remaining: The estimated remaining useful life of the worst-case SSD.
- SSD Location: The slot and node of the worst-case SSD.
- SSD Confidence: The degree of certainty regarding the worst-case SSD's RUL.
SSD life expectancy is based on survival analysis. To view the SSD Health forecast for a system, click the Analytics window and select the source.
The Economy Storage forecast takes advantage of the storage usage value and total storage available value to predict up to a year’s worth of storage needs. Additional to the forecasted values is a confidence interval (also known as a cone of confidence) that extends up to a year.
These forecasts are generated daily and take into account both weekly and monthly seasonal trends to help further articulate their predictions. The data is stored and collected every three minutes and is displayed on the Analytics dashboard.
The following information is displayed on your forecast:
- Days Remaining: Estimated days until capacity is reached.
- Capacity: Total storage capacity.
- Confidence: The degree of certainty that the actual storage used will fall within the estimates.