Prism Central includes machine-learning capabilities that analyze resource usage over time and provide tools to monitor resource consumption, identify abnormal behavior, and guide resource planning. These tools include
-
VM "right sizing" where VMs are analyzed and those that exhibit inefficient profiles are identified.
-
Anomaly detection to record when performance or resource usage is outside an expected range based on learned VM baseline behavior.
-
"Smart" alerts that trigger when specified anomalies are recorded.
-
Reports that summarize cluster efficiency.
VM Right Sizing
It is useful to look at the profile of your VMs when analyzing problems in a cluster or assessing future resource needs. This can help you identify VMs that are not optimally configured such as ones that consume too many resources, are constrained, are over provisioned, or are inactive.
Anomaly Detection:
The right sizing feature identifies inefficient VMs that fit one of the profiles described as below:
-
Bully VM : A "bully" VM is one that consumes too many resources and causes other VMs to starve.
-
Constrained VM : A "constrained" VM is one that does not have enough resources for the demand and can lead to performance bottlenecks.
-
Over- provisioned VM : An "over-provisioned" VM is the opposite of a constrained VM, meaning it is a VM that is over-sized and wasting resources which are not needed.
-
Inactive VM : A VM is "inactive" in either of the following states:
"Dead" VM: A VM is considered dead when it has been powered off for at least 21 days.
"Zombie" VM: A VM is considered a zombie when it is powered on but does fewer than 30 read or write I/Os (total) and receives or transfers fewer than 1000 bytes per day for the past 21 days.
The system predicts a normal behavior band for various metrics based on historical data. The anomaly detection module monitors a predefined set of metrics on a daily basis and publishes baseline values for each of the metrics.
-
Twenty-seven metrics are monitored for VMs, hosts, and clusters.
-
Data for each metric from the past 21 days is recorded and analyzed, a normal behavior band is established, and predictions for the next 2 days are formulated.
-
The behavior bands and predictions are adjusted accordingly when time period or trend patterns are observed, for example low CPU on weekends or increasing CPU usage.
Smart Alerts
You can create custom policies to generate alerts when behavioral anomalies occur. You can generate a critical or warning alert when a behavioral anomaly occurs for the following conditions:
-
a specified VM, host, or cluster
-
a specified metric
-
every time or only when the anomaly resides within a certain range
Cluster Efficiency Reports
This is one of the default reports.This report provides cluster statistics (host and VM counts), performance and usage statistics (average and peak CPU and memory usage), runway metrics, and information about inefficient VMs (bully, constrained, over-provisioned, and inactive). You can run this report to get a quick view of how efficiently the cluster is performing.
For details about these individual tools, please refer to the portal documentation.
Portal documentation: https://portal.nutanix.com/page/documents/details?targetId=Prism-Central-Guide-Prism-v5_19:mul-behavioral-learning-pc-c.html