|Partner TechTip: Collecting Your Hidden Performance Data|
|Tips & Techniques - System Administration|
|Written by Tom Huntington|
|Friday, 07 September 2012 00:00|
Get continuous performance data without major overhead.
Does your team often get surprised by end users calling to complain about performance problems? Have you ever not had the details needed to see what was causing a performance problem? Are you struggling with a proactive approach to managing performance? Do you have someone manually documenting WRKSYSSTS or WRKDSKSTS output?
Think Robot/NETWORK Explorer.
On IBM i, performance doesn't have to be an issue: The operating system gives you the ability to continuously collect performance data—without major overhead on the system.
IBM i has a feature called Collection Services, a tool set designed by IBM to gather the performance data on IBM i and store it in intervals inside a collection object (object type *MGTCOL). These objects store the raw performance data that reflects an interval of time and the performance of the system, jobs, memory pools, and disk.
This data is essential for a few good business reasons. It can be used to diagnosis performance issues, plan for future growth, or monitor performance for exceptions before a problem occurs. The command CRTPFRDTA processes the data in the *MGTCOL object and outputs it to performance database files where you can harvest the data with a high-level program or use some other IBM tools—like performance tools—to look at the data.
The STRPFRCOL command allows you to start the process that monitors the data and controls the intervals for taking snapshots of the data. IBM now recommends running performance collections constantly as part of normal operations.
Performance collection data can be evoked and run through system APIs, as well as through IBM i Navigator and these commands. We use APIs in many software solutions at Help/Systems. It keeps us performing at an optimal speed, too.
Figure 1: This is an example of "Active to Ineligible" values.
Management by Exception = SLA Delivered
Managing your server by exception is especially important for performance data, and Robot/NETWORK makes it easy to add exception-based monitoring. It lets you configure elements of performance by each partition, allowing you to choose only important items from WRKSYSSTS or WRKDSKSTS. For instance, you might want to get alerted if you go over a threshold of .05 seconds for average response time or if your active to ineligible value is greater than 5 (Figure 2). This is very easy to do, by partition or across IBM i.
Figure 2: The Status Center shows events as you exceed your desired threshold.
Robot/NETWORK also features auto-refreshing dashboards (Figure 3) to keep you apprised of your most recent data. The product can also alert you for each interval collection, which can also be summarized and compared to the threshold. Threshold events go to the status center, which in turn can be set to notify you via SNMP, email, pop-up window, or desktop alert.
Figure 3: The Robot/NETWORK dashboard is part of the product's new Performance Center.
Meet Your SLA
|Last Updated on Friday, 07 September 2012 00:00|