Performance data centralization: HMC or IVM

On systems managed by HMC or IVM it is possible to centralize performance data collection. This feature is not enabled by default. The HMC and IVM collect the following types of utilization data: sampling events, state change events, and configuration change events.

Sampling events are collected for the managed system, for each partition in the managed system, and for the shared processor pool in the managed system. Events are collected hourly, daily at midnight, and monthly at midnight on the first day of each month. Information is related to memory and processor utilization.

State change events are collected for the managed system and for each partition in the managed system. They are collected when a state change occurs for the managed system or a partition.

Configuration change events are collected for the managed system, for each partition in the managed system. They are collected when a configuration change affecting memory or processor resources occurs for the managed system or for a partition.

Hourly sampling events and all state change and configuration change events collected are only saved on the HMC or IVM for about 2 months. Daily sampling events are saved for about 2 years, and monthly sampling events are saved for about 10 years.

Data can be retrieved using the lslparutil command on the HMC or IVM and stored on the system that has pGraph installed. Since both HMC and IVM are appliance servers with a closed environment, a typical scenario is to have an external system that connects data using SSH, issues the lslparutil command and stores locally the output on a file.

For example, if a user can remotely execute command on the HMC, performance data can be collected in the following way:

> ssh -l hscroot HMC.mycompany.com lslparutil -r all -m ManagedSystem1 -d 10 > hourly-10days.output

> ssh -l hscroot HMC.mycompany.com lslparutil -r all -m ManagedSystem2 --startyear 2007 -s d > daily-from-Jan2007.output

The first example extracts data related to ManagedSystem1 with hourly samples related to the last 10 days, while the second extracts data related to ManagedSystem2 with daily samples from the beginning of year 2007.

pGraph currently handles only CPU data collected by HMC: global usage, free pool status and single LPAR consumption.

An example of pGraph output is in the following figure. An HMC has been collecting performance data from a p595 for more than two months and daily CPU usage has been plotted by pGraph. Only a subset of available graphs has been shown. The six graphs represent:

1. The number of LPARs defined in the system.
2. Global CPU usage split by LPAR.
3. For only micro LPARs, the sum of entitlements, virtual processors and processor consumed.
4. System pool status (size and free pool).
5. Physical CPU consumption by LPAR, related to entitlement and virtual processors.