The z13 processor range comes with very large real storage configurations with 1 Mbyte and 2Gbyte pages. Despite all the improvements that have been made, virtual storage below the 16 Mbyte line is still scarce, and lack of it can still cause an unplanned outage. IntelliMagic Vision checks the health of real and virtual storage configurations and catches problems before they impact LPAR availability

Effective use of Large Frame Area and TLB

It is up to the installation to configure the Large Frame Area (LFAREA), which is the area containing large 1Mbyte and 2Gbyte pages. You need to ensure effective use of the LFAREA storage that is defined, and make sure that the demand for 1 MB pages can be met.

Large pages can improve the effectiveness of the Translation Lookaside Buffer (TLB), which will give you a higher MIPS rate at no cost. This is because with large pages fewer unique pages need to be referenced, and hence the TLB will need fewer entries to run each program.

IntelliMagic Vision shows all the relevant statistics for TLB usage and TLB related delays and assesses whether they indicate a healthy situation or not by computing a ‘rating’ that is green, yellow or red (see example below). These TLB efficiency reports are provided in the Processor Reporting Report Set, based on CPU Measurement Facility data (SMF 113 records).

Screenshot of processor hardware counters by processor complex serial
MIPS Cost of TLB mechanism: about 3.7% of the processor capacity is ‘lost’ because of the TLB misses. This is considered a good result, hence IntelliMagic Vision rates this situation as ‘healthy’, with a green frame.

Analyzing the impact of paging

The Real Storage Manager (RSM), uses paging to manage the demand for central storage frames.

As large memory buffers are used by DB2, Java and applications, the pressure on central storage will increase. It is important to make sure that no significant paging occurs, particularly no demand paging for server address spaces such as DB2, IMS, and CICS, and no paging in general while taking system dumps.  It is also important that sufficient page data set and SCM (“flash memory”) space is available for paging. Paging problems seriously impact performance as the latency required to retrieve pages from DASD is measured in milliseconds as opposed to nanoseconds for main storage retrieval. The paging dashboard in IntelliMagic Vision monitors whether there are any paging problems or errors that should be addressed. This dashboard should always be completely green. Green means there is healthy central storage and paging environment: high UIC, little paging, and enough space on the page data sets.

When paging occurs, you will want to investigate:

As large memory buffers are used by DB2, Java and applications, the pressure on central storage will increase. It is important to make sure that no significant paging occurs, particularly no demand paging for server address spaces such as DB2, IMS, and CICS, and no paging in general while taking system dumps.  It is also important that sufficient page data set and SCM (“flash memory”) space is available for paging. Paging problems seriously impact performance as the latency required to retrieve pages from DASD is measured in milliseconds as opposed to nanoseconds for main storage retrieval.

The paging dashboard in IntelliMagic Vision monitors whether there are any paging problems or errors that should be addressed. This dashboard should always be completely green. Green means there is healthy central storage and paging environment: high UIC, little paging, and enough space on the page data sets.

When paging occurs, you will want to investigate:

  • which workloads suffer from paging,
  • which workloads use most of the storage.

This can be done with the RMF 72 Workload Manager information that provides both paging and storage occupancy at the service class level. IntelliMagic Vision visualizes this information in a way that makes it possible for you to easily analyze both the impact and the cause of paging. Below is an example, taken from a multichart that also shows CP, zIIP, and zAAP usage, as well as the I/O activity.

IntelliMagic Vision provides insight in how your workloads may be impacted by paging, and how they use central storage. These particular charts provide the sysplex totals for paging and page residency summarized by workload.

Watch the fixed storage below the line

The current 64-bit z/OS architecture evolved from earlier 24-bit (MVS) and 31/32-bit (MVS/XA) versions. The 16 Mbyte memory addressable by the 24-bit version (MVS) is called the memory ‘below the line’. The 2 Gbyte addressable memory from the 31/32-bit version (MVS/XA) is called ‘below the bar’. This legacy is still important, since certain z/OS functions only work below the 16 Mbyte line, and others need storage below the 2 Gbyte bar. These limitations are mostly virtual storage issues, but systems that fix many frames below the line typically see a higher than average page movement rate (SMF71PMV), a process where RSM needs to shuffle pages between the areas above and below the 16 Mbyte line, generating system overhead. With IntelliMagic Vision you can watch the fixed storage below the 16 Mbyte line.

Picture of maximum fixed frames in central storage below 16 MB line.
Real storage frames below the line that are fixed in memory. Best practice is to have no more than 2 Mbyte fixed.

Monitoring the Pageable 1Mbyte frames

z/OS will allocate about 1/8th of the real storage to pageable 1 Mbyte frames. If the number of 1 Mbyte pages requested by applications will not fit in that originally allocated space, z/OS will create them in the 1 Mbyte fixed LFAREA, or assemble them from 4 Kbyte pages.

The following chart for unused frames is rated by IntelliMagic Vision, to let you know when these frames are exhausted. This means that the requests will be served from the LFAREA or the 4 Kbyte areas. When the LFAREA is also full, you should increase the size of the LFAREA to provide more 1 Mbyte frames.

Usage of pageable 1 Mbyte frames for the LPARs in a sysplex. Note that this installation does not see much use of the 1 Mbyte pageable frames, less than 10% for each of the LPARs.

Technical Information on IntelliMagic Vision for z/OS

Related resources

Blog

IBM z15 Announcement Highlights and How to Take Advantage

The z15 (with a General Availability date of 9/23/2019) offers up to 190 CPU cores (vs. 170 on z14) and 40 TB of usable memory (vs. 32 on z14), in addition to processor cache and overall performance improvements.

Read more
Webinar

Top Performance Strategies for Black Friday, Cyber Monday, and other Peak Periods

Learn about performance strategies to handle your peak periods, like Black Friday and Cyber Monday, so that your infrastructure is able to deliver the required availability.

Watch Webinar
Webinar

How to Learn (Almost Anything) from RMF & SMF Data

Learn how to gain a vision for how rapidly you could leverage easy visibility into RMF and SMF data to expand your understanding of how the z/OS ecosystem operates.

Watch Webinar

Go to Resources