Identifying remaining performance capacity
Performance capacity, or headroom, measures how much work you can place on a node or an aggregate before performance of workloads on the resource begins to be affected by latency. Knowing the available performance capacity on the cluster helps you provision and balance workloads.
Before you begin
Advanced privilege level commands are required for this task.
About this task
You can use the following values for the -object option to collect and display headroom statistics:
For CPUs, resource_headroom_cpu.
For aggregates, resource_headroom_aggr.
You can also complete this task using ThinkSystem™ Storage Manager for DM Series and Unified Manager for DM Series.
Example
The following example displays the average hourly headroom statistics for cluster nodes.
You can compute the available performance capacity for a node by subtracting the current_utilization counter from the optimal_point_utilization counter. In this example, the utilization capacity for CPU_sti2520-213 is -14% (72%-86%), which suggests that the CPU has been overutilized on average for the past hour.
You could have specified ewma_daily, ewma_weekly, or ewma_monthly to get the same information averaged over longer periods of time.
sti2520-2131454963690::*> statistics show -object resource_headroom_cpu -raw -counter ewma_hourly
(statistics show)
Object: resource_headroom_cpu
Instance: CPU_sti2520-213
Start-time: 2/9/2020 16:06:27
End-time: 2/9/2020 16:06:27
Scope: sti2520-213
Counter Value
-------------------------------- --------------------------------
ewma_hourly -
current_ops 4376
current_latency 37719
current_utilization 86
optimal_point_ops 2573
optimal_point_latency 3589
optimal_point_utilization 72
optimal_point_confidence_factor 1
Object: resource_headroom_cpu
Instance: CPU_sti2520-214
Start-time: 2/9/2020 16:06:27
End-time: 2/9/2020 16:06:27
Scope: sti2520-214
Counter Value
-------------------------------- --------------------------------
ewma_hourly -
current_ops 0
current_latency 0
current_utilization 0
optimal_point_ops 0
optimal_point_latency 0
optimal_point_utilization 71
optimal_point_confidence_factor 1
2 entries were displayed.