Analyzing processor usage

Processor busy alone does not sufficiently indicate processor capacity requirements. You should investigate high processor busy to determine which workloads are causing it. Having a processor at 100% busy indicates that processor capacity is at a maximum, and adding workload will not provide more answers. Lower priority work waits longer for processor cycles. This may or may not be acceptable, depending on your service-level and response-time requirements.

Processor performance issues occur when a lack of processor cycles causes service-level objectives to be missed. This can occur when:

  • High priority work dominates the processor
  • An application runs at the speed of a processor
  • A lot of work arrives at the same time (workload arrival patterns)

You must analyze low processor busy to determine if another system resource is causing response-time delays. Processor contention is demonstrated by showing processor busy indicators and latent demand indicators.

The average processor busy for the shift is the average busy of all the processors in the system. In a multiprocessor system, if each processor was 70% busy for the interval, then the average system busy is 70%.

In an LPAR environment, the processor busy represents the logical busy per processor based on the logical dispatch time.

IBM Z Performance and Capacity Analytics provides a number of reports that show CPU usage. The MVSPM Average CPU Busy, Hourly Trend report presents the hourly usage trend for your system. This report includes all systems for which data is collected.

Figure 1. MVSPM Average CPU Busy, Hourly Trend report

                      Sysplex: 'SVPLEX5 '  Date: 2006-05-13
                                   Period: All
                               Processor type: CP

                        MVS_system
                        <- S5C -->      < TOTAL ->
   Time
   -----              --------------  ---------------
   13:00                    9.7             9.7
   14:00                   74.5            74.5
                    ==============  ===============
                           42.1            42.1



      IBM Z Performance and Capacity Analytics: MVSPM07

This report presents the overall picture of your processor usage for the day you specified and for the processor type (CP, IFA or IIP) you specified when creating the report. Processor busy patterns are an important part of the performance road map. Processor contention problems occur at peak periods. You can easily identify which times and which system to investigate for more detail. You can investigate the busy systems to look for delays caused by insufficient processor capacity. You should also look at the systems that seem to be underused. Under utilization is not always caused by a lack of work. Sometimes the processor cycles cannot be used because of a resource contention, such as storage constraints and delays in transferring data. At times, you must look at the impact of other systems, particularly in the area of DASD sharing or system consolidation.

The arrival time of work on different systems is an important factor in processor load balancing. When planning to use LPAR as a vehicle for processor load balancing, this factor helps determine if processor balancing will be beneficial.