z/OS storage performance glossary: the key terms of RMF, CMF, DCOLLECT and tape

This z/OS storage performance glossary defines, in one sentence each, the terms an engineer meets when reading RMF, CMF, DCOLLECT and tape data. It is grouped by subject: measurement sources, disk, CPU, tape and space, with a table of where each term is measured.

Storage performance has its own vocabulary, and the same word can mean different things in different reports. The definitions below follow IBM usage. Use the headings to jump to a subject.

Measurement sources

SMF (System Management Facilities) is the z/OS component that collects and writes measurement and accounting records. Each record has a type and often a subtype, such as 74.1 for device activity.

RMF (Resource Measurement Facility) is the IBM performance monitor of z/OS. Its Monitor I writes SMF records types 70 to 79 at each interval.

The RMF Post Processor is the batch program, ERBRMFPP, that turns RMF records into printed reports. Each report covers one subject, such as devices, cache or CPU.

CMF MONITOR is the BMC performance monitor for z/OS, an alternative to RMF. Its CMF Analyzer prints reports equivalent to those of RMF, with other layouts.

An interval is the period over which RMF or CMF accumulates measurements before it writes a record, commonly fifteen or thirty minutes. All averages in a report are per interval.

DCOLLECT is an IDCAMS command that writes data set, volume and policy information to a data set. It is the usual source for space analysis.

VEHSTATS is an IBM Tape Tools utility that formats the historical statistics of a TS7700 into reports. BVIR (Bulk Volume Information Retrieval) is the TS7700 function that returns those statistics to the host.

Disk terms

DASD (direct access storage device) is the z/OS name for disk storage. A DASD volume is identified by a six-character volume serial (volser).

Response time is the average time an I/O to a volume takes, from the request in z/OS to the end of the transfer. RMF splits it into four components:

• IOSQ time is time spent queued in z/OS because the device is already busy with a request from the same system.
• Pending time is time between the start of the I/O and its acceptance by the channel path, control unit and device.
• Disconnect time is time the storage system works on the I/O while the channel is free, for example to read data from the back end.
• Connect time is time spent transferring commands and data over the channel.

An LCU (logical control unit) is a group of device addresses that z/OS addresses as one control unit. An SSID (subsystem identifier) identifies a group of volumes in the storage system, and cache statistics are reported per SSID.

A PAV (parallel access volume) is a volume reachable through a base address and alias addresses, so that several I/Os can run against it at once. HyperPAV assigns aliases from a pool per LCU to each I/O, and SuperPAV lets LCUs of the same group share their aliases.

The cache hit ratio is the share of I/Os served from the storage system cache. NVS (non-volatile storage) is the protected memory where the storage system keeps writes until it destages them to the drives.

FICON is the channel protocol that connects IBM Z to storage systems. zHPF (High Performance FICON) is a FICON extension that runs channel programs with less overhead. A CHPID (channel path identifier) is the number of a channel path in the channel subsystem.

An extent pool is a set of ranks from which a DS8000 allocates volumes. A rank is a RAID array of drives in a DS8000.

CPU terms

An LPAR (logical partition) is a share of a physical machine that runs its own operating system image. A CPC (central processor complex) is the physical machine that hosts LPARs.

An MSU (million service units per hour) is the unit z/OS uses for processor capacity and LPAR consumption. MIPS is an informal throughput figure that z/OS does not report.

The R4HA (rolling four-hour average) is the average MSU consumption of an LPAR over the last four hours. A defined capacity is an MSU limit on an LPAR; WLM caps the LPAR when the R4HA exceeds it.

An LPAR weight is the share of the shared processors that PR/SM guarantees to an LPAR when the processors are busy.

WLM (Workload Manager) is the z/OS component that manages work against goals set in a policy. A service class is a group of work with the same goal and importance. A report class is a group of work used only for reporting. The performance index is the ratio between the goal and what the work achieved; above 1, the goal was missed.

A zIIP is a specialty processor for eligible work, counted apart from general purpose processors.

Tape terms

A virtual tape system stores tape volumes on disk and presents them to z/OS as tape drives and cartridges. The IBM TS7700 is such a system.

The tape volume cache (TVC) is the disk cache of a TS7700. A recall is the copy of a virtual volume from physical tape back into the cache. Throttling is a delay the TS7700 adds to host writes to protect its cache or its copies.

A scratch volume is a tape volume with no active data, free for a new write. An active volume holds data that is still kept.

A tape management system, such as DFSMSrmm, CONTROL-T or CA-1, keeps the catalogue of tape volumes and their data sets.

Mount pending is the state of a tape device between a mount request and the moment the volume is ready.

Space terms

A storage group is a set of volumes that SMS manages as one pool for allocation. SMS (storage management subsystem) is the DFSMS component that places data sets according to policy.

Free space is the space on a volume not allocated to any data set. The largest free extent is the biggest contiguous free area on a volume. The fragmentation index is a number that grows as free space is split into more, smaller pieces.

An extent is a contiguous area of a volume allocated to a data set. Defragmentation is the consolidation of free space into fewer, larger extents, for example with DFSMSdss.

Where each term is measured

How zPulse uses these terms

zPulse reads the printed RMF and CMF reports, DCOLLECT extracts, tape catalogue reports and VEHSTATS files listed above. It reads no raw SMF records. Its screens use the same vocabulary: the four response time components, LCU, SSID, HyperPAV, MSU, R4HA, service class, storage group and fragmentation index.

The zPulse tools page lists the 25 tools with the file each one reads, and the about page tells how Technical Storage has worked with these reports since 1997.

To see these terms applied to your own systems, request a free study.