The RMFSRV query displays resource usage performance indicators, summary sampled state percentages, and detailed execution delay state information for each Workload Manager (WLM) Service Class Period. This information is provided by the following three data extract CSV’s at the z/OS LPAR (SYSID), Central Processing Complex (CPC), and SYSPLEX levels:
Each data extract generates the following four chart views:
The title area for all RMFSRV query charts shows the SYSPLEX name, Service Class Name, Relative Importance, Goal Type, and Service Class period charted. As delivered, each RMFSRV query chart shows the hourly activity for the past two days. For each SYSPLEX, the Service Class Periods are presented in Relative Importance order, with the most important Service Classes presented first.
Chart view #1, CPU and PI, the general processor CP, zIIP, and zAAP CPU time used by the Service Class Period (SCP). The CPU times used for each processor engine type are stacked and are scaled using the left Y-axis. Two lines are plotted against the right Y-axis scale - the Performance Index (PI) for the SCP, and a Performance Index "goal met" reference line, set to 1.
To prevent scaling issues, PI values greater than 5 are set to 5. PI values between zero (0) and one (1) mean that the Service Class exceeded the specified performance goal. A value equal to 1 means that the goal was met exactly. Values greater than 1 means that the specified performance goal was not met. The PI cannot be computed for certain goal types such as DISCRETIONARY and SYSTEM SPECIFIED.
The first chart sample was charted from the data extract: Service Class by SYSPLEX CPCUID.Shown is a two day hourly chart for Period 1 of Service Class STCTEST. This Service Class executed in SYSPLEX PLEXC1, and the information displayed occurred on z/OS LPARS executing on an IBM 2817-615 CPC.
Workloads in the STCTEST Service Class use both CP and zIIP processor engines, as shown by the blue and red stacked bars. Service Class STCTEST has an execution velocity goal type. The actual performance index value, shown with the beige line, is generally below the reference value of 1, meaning that the actual performance of workloads running in this Service Class generally exceeded the specified goal. But from 5 PM to 7 PM on the first day, the PI was value was 5, meaning the goal was severely missed.
Chart view #2, Sampling Summary, displays the percentage of times workloads for the SCP were found to be in one of three states: Using, Wait-Unknown, and Execution Delays. The percentage of each state is stacked using vertical bars and are scaled using the left Y-axis fixed to show values from 0 to 100 percent. A fourth state which is not shown, if added to the chart would bring the stacked bars up to the 100% line, is the percentage of samples where workloads in the SCP were idle.
The Using state represents the percentage of samples where the workloads were using the CPU, and if I/O Priority is active, performing I/O. The Wait-Unknown state represents the percentage of workload delay states that WLM does not manage. These delays include time waiting on an I/O, DB2 latches, tape mount, enqueues, and others. The Execution Delay state represents the percentage of samples where the SCP workloads experienced execution delays for resources that WLM does manage.
Two lines are plotted against the right Y-axis scale—the performance index (PI) for the Service Class period, and a performance index "goal met" reference line, set to 1.
Chart views #3 and #4, Exec Delay States Group 1 and Exec Delay States Group 2, show the percentage of samples belonging to the various categories of execution delays. There are too many categories to effectively display using a single chart, so two chart views are provided, Group 1 and Group 2. The sample percentages of each delay category are stacked using vertical bars against the left Y axis. The axis scale is dynamically adjusted based on the percentages encountered in the data. As with the other chart views, two lines are plotted against the right Y-axis scale—the performance index (PI) for the Service Class period, and a performance index "goal met" reference line, set to 1.
The second chart sample above was displayed by selecting view #3, Exec Delay States Group 1. This view shows the first set of execution delay category percentages for Period 1 of the Service Class STCTEST. You can see that the reason the SCP goal was missed for hours 5 PM through 7 PM on the first day was due to an increase in delays for the STCTEST workloads while waiting to be dispatched on a CP processor.
|
Copyright © 2014 CA.
All rights reserved.
|
|