Todd Havekost - 29 June 2022

Following up on my first blog covering integrated visibility into WLM and CICS data, this blog continues our series demonstrating how mainframe performance analysis is enhanced by having integrated visibility into SMF data across data types. In this blog I’ll demonstrate how address space (SMF 30) and Db2 Accounting (SMF 101) data complement one another and provide a more complete “big picture” view than is available from either data source on its own.

Infrastructure teams today commonly face the barrier of learning different tooling if they want to collaborate with other teams and view SMF data types created by other components. This leads to the undesirable outcome of not having access to insights that are available from other SMF data types. Instead, teams are “handcuffed” by having to hand questions off to other areas to analyze other SMF data types using tooling specific to that discipline. This obstacle can be overcome through a common, intuitive, integrated interface that gives all teams visibility into the insights provided by components across the entire z/OS platform.

Address Space (SMF 30) and Db2 Accounting (SMF 101) data each provide useful insights necessary for data analysis, and in some use cases may be all you need. However, in certain situations, such as trying to understand Db2 batch job activity, gaining a complete “big picture” view often requires visibility provided from both data sources.

Scenario 1: Address Space Perspective (SMF 30)

One “workhorse” source of insights into the operation of a z/OS system comes from SMF 30 records that provide data by address space at job, interval, or step levels. But for Db2 batch jobs, one challenge with only having this view is that all the activity is rolled up into a step record with PGM=IKJEFT01. This has valuable data for many z/OS metrics such as CPU and I/O, but gives little idea of the nature of the Db2 activity that is taking place (e.g., getpages, SQL statements).

Figure 1 shows a list of top jobs by total CPU consumption that are executing PGM=IKJEFT01.

Processor Time for Address Space names

Figure 1: Processor Time (Top 10)

 

This excerpted view of the step termination records (sorted by descending CPU usage) for a selected job shows that most of the CPU and elapsed time occurs in two of the more than 100 steps (Figure 2).

Job Step Termination Records (excerpt)

Figure 2: Job Step Termination Records (excerpt)

 

If the z/OS team is limited to viewing SMF 30 address space data, that is the extent of what can be determined about this job. But if the team also has easy access to view Db2 accounting data, many more insights become available.

Scenario 1: Db2 Accounting Perspective (SMF 101)

Db2 Usage drill down

Figure 3: “Db2 Usage” drill down

Figure 3 illustrates how analysis can proceed directly (from Figure 1) by leveraging the fact that for several connection types, including Db2 Call Attach which is commonly used by jobs, Db2 captures the z/OS job name in the correlation name field.

Matching up the Db2 accounting data for this job enables new potential analysis into Db2 activity, including logging, row activity, getpages, and prefetch requests as seen below in Figure 4.

 

Plan Overview summary view of several Db2 metrics

Figure 4: “Plan Overview” summary view of several Db2 metrics

 

Selecting and expanding the view of SQL Statements indicates that these jobs are generating approximately 1000 SQL Fetch commands per commit and an equivalent number of SQL Delete commands (Figure 5).

SQL Statements

Figure 5: SQL Statements

 

This illustrates how Db2 accounting (Db2 101) data can provide extensive insights on what is happening within Db2 that is not available from address space (SMF 30) data.

Scenario 2: Db2 Accounting Perspective (SMF 101)

But Db2 accounting data has its own limitation. It does not generate interval data, but instead its data is only written at thread termination. This means that long-running work may only generate a single set of accounting metrics at end of job, providing no insights into the level of activity across the life of the job.

Starting from a view of top CPU consuming Db2 Call Attach jobs, a user may want to identify activity over time for a selected job (Figure 6).

Db2 and Application CPU Usage (Top 10)

Figure 6: Db2 and Application CPU Usage (Top 10)

 

Since accounting data is written only at thread termination, the resulting view (Figure 7) reports all data at end of job and provides no visibility into the level of activity over the more than one hour duration of the job.

Db2 and Application CPU Usage for selected job

Figure 7: Db2 and Application CPU Usage for selected job

 

Scenario 2: Address Space Perspective (SMF 30)

Figure 8 illustrates how analysis might proceed by moving from the Db2 accounting data into SMF 30 address space data where its interval recording can fill in this gap.

Db2 and Application CPU Usage (Top 10)

Figure 8: Db2 and Application CPU Usage (Top 10)

 

Indeed, the SMF 30 interval data comes to our rescue, indicating steady levels of GCP and zIIP CPU over the duration of the job, as well as increasing time consumed by disk I/O operations in latter time intervals (Figure 9).

Address Space Interval Statistics

Figure 9: Address Space Interval Statistics

 

Analysis Enhanced through Viewing Multiple Types of SMF Data

This use case illustrated how having a common interface across data types enhanced analysis by enabling z/OS and Db2 teams to gain a more complete picture of the activity of Db2 jobs by having easy access to both address space (SMF 30) and Db2 accounting (SMF 101) data.

Subsequent blogs in this series will provide examples of other real-life scenarios where visibility into insights derived from multiple types of SMF data enhances analysis by providing a more complete perspective into the operation of a z/OS environment.

The below video demonstrates the navigation involved in creating these views using IntelliMagic Vision for z/OS.

 

Related Resources

Whitepaper

Learning from SMF: Db2 Buffer Pools

Dive into the SMF type 100 records to see what insights they offer into Db2 buffer pools. This whitepaper contains valuable information for relative newbies and experienced hands alike.

Download
Webinar

zAcademy: zHyperLink Read Performance & Lessons Learned

Watch Joe Hyde as he shares recent Db2 production data which highlights the speed of zHyperLink read I/Os and other lessons learned.

Watch Webinar
Webinar

zAcademy: Insights You Can Gain from Integrated Visibility Across Types of SMF Data

Expand your awareness of ways your analysis can benefit from combining insights across z/OS data types.

Watch Webinar

Go to Resources

This article's author

Todd Havekost
Senior z/OS Performance Consultant
More from Todd

Share this blog

Troubleshooting WLM Missed Goals with CICS Transaction Data

Subscribe to our Newsletter

Subscribe to our newsletter and receive monthly updates about the latest industry news and high quality content, like webinars, blogs, white papers, and more.