Measuring processor time

Processor time is a measure of the time that a job or task controls the processor. When a job receives control of the processor, the operating system stores the current time-of-day (TOD) clock value. When the job loses control of the processor, z/OS subtracts the current TOD clock value from the start value, and records the results as the duration of the job.

The primary sources of z/OS processor usage are:

SMF type 70 record

This record provides total system-wide processor usage for a recording interval. An SMF type 70 record includes the number of seconds in the interval that the processor was in wait state (that is, waiting to execute instructions). When creating this record, RMF does not use sampling. RMF uses the hardware processor TOD clock as its source for the processor wait time for the measured interval.

The processor active time is the duration of the interval minus the wait time. The ratio of processor active time to interval duration, expressed as a percentage, is the common measure of percent processor busy.

Many processor complexes consist of several individual processors that share system resources and work. The operating system that manages the work that is using these processors has access to all processors for both work and measurement purposes. RMF supports all processors that are physically or logically made available to the operating system. RMF stores the wait times for each processor in the same SMF type 70 record at the end of each measurement interval.

If a processor complex includes four individual processors all physically available to the z/OS system, then four measures of processor wait time will be found in each record at the end of each interval.

A processor can be logically made available through the Processor Resource/Systems Manager (PR/SM™).

For an LPAR partition with wait completion enabled, wait time is subtracted from logical processor dispatch time. For an LPAR partition with wait completion not enabled, wait time is not used. Logical processor dispatch time is used.

SMF type 72 record

This record provides processor usage data for a performance group for the recording interval. You can subtract the sum of the performance group processor times from the total processor active time derived from an SMF type 70 record to measure unassigned processor time. This unassigned time is the processor active time that z/OS does not directly associate with any performance group.

Processor time includes:

TCB time

The time a job executes under a task control block (TCB).

SRB time

The time a job executes under a service control block (SRB).

RCT time

The time used for a region control task (RCT).

IIT time

The time used for processing I/O interrupts (IIT).

HST time

The time used for hiperspace service (HST).

z/OS records these processor times separately in SMF type 72 records.

z/OS records the SRB and TCB times as service units. z/OS converts each of the times into service units by multiplying the time by an SRM processor-dependent constant. For example, z/OS converts the time a job executes under a TCB into TCB service units by multiplying the TCB time by an SRM constant that is processor-dependent. RMF adds the value to the TCB service unit field of the SMF type 72 record of the performance group that the job belongs to. z/OS converts the SRB time in the same way.

z/OS records RCT time in 0.001 024 seconds, and IIT and HST in 0.000 001 seconds.

Starting from z/OS version 1.6, the SMF type 72 record also contains:

IFA Time

The CPU time spent on the Integrated Facility for Application (IFA) coprocessor. IFA Time, in hundredths of a second, is present as a separate field, but it is also added to the TCB time if the IFA eligible work runs on the standard CPs.

Starting from z/OS version 1.8, the SMF type 72 record also contains:

IIP Time

The CPU time spent on the Integrated Information Processor (IIP) processor. IIP Time, in hundredths of a second, is present as a separate field, but it is also added to the TCB time if the IIP eligible work runs on the standard CPs.

Report performance groups are also recorded in SMF type 72 records. Report performance groups report statistics on transactions. Depending on the number of ways statistics are collected, there can be more than one report performance group for a transaction. Exclude report performance groups when summaries are produced to avoid counting workload more than once.

The SMF type 72 record contains subtypes 1 and 2 when z/OS is running in compatibility mode, and types 3 and 4 when z/OS is running in goal mode. Subtypes 1 and 2 contain information about performance groups, and subtypes 3 and 4 contain information about service classes.

SMF type 30 records

This record provides information about processor usage at the job level. A type 30 record contains operation information, such as the job and step start and end times, step processor time, step termination status, device allocation start time, and problem program start time, for a job or step.

In this record, z/OS records processor time as TCB, SRB, RCT, IIP (processing I/O interrupts, same as IIT in SMF type 72 record), HPT (same as HST in SMF type 72 record), and IFA & IIP (CPU time spent on Integrated Information Processors), in hundredths of a second. This record also provides initiator processor time.

Each job or step is associated with a performance group number. You cannot compare data in SMF type 30 records to SMF type 72 records, because they do not show processor usage for the same time period. SMF type 30, subtypes 4 and 5, records show total CPU used for the step and job.

Figure 9 illustrates the division of processor times and the overlap of the different records.

Figure 9. Processor times

Understanding capture ratio

Figure 9 shows that not all of the processor time captured in SMF type 70 records is captured in SMF type 72 records. For example, the processor time consumed in the termination of the job or step and the time used to write the job or step record to SMF is not included in the SMF type 72 record, but is included in the SMF type 70 record.

The capture ratio is the ratio of accounted-for time in the SMF type 72 record to the total processor time measured in SMF type 70 record. You compute the capture ratio by comparing the data in the SMF type 72 record and the SMF type 70 record.

Several things can make the capture ratio vary. For example, a storage-constrained system will show a lower capture ratio. A small RMF cycle time (below 1 second), large dispatching queues, and implementation of SLIP PER trap activity cause variations in the capture ratio. What can cause the capture ratio to vary depends on the release of z/OS being used. Refer to System Management Facilities (SMF) for the release of z/OS that you are using.

[ Top of Page | Previous Page | Next Page | Contents ]