From one system to many: sysplex, XCF and couple data sets
A sysplex is a group of z/OS systems that cooperate as one. XCF lets them send messages to each other and notice when one fails, couple data sets hold the shared records that keep them in agreement, and a common time source keeps their clocks aligned.
Why one system is not enough
A single z/OS image in an LPAR can be very large, but it is still one point of failure, and every planned IPL for maintenance takes its work down with it. A sysplex (short for systems complex) joins up to 32 z/OS images so they can share work and cover for each other. The images can sit on one CPC or be spread across several machines, which is better for availability.
When people at work say "the sysplex" they almost always mean a Parallel Sysplex. A single system running with sysplex services but no partners is sometimes called a monoplex; you will see these on small test systems.
XCF: how members talk to each other
The cross-system coupling facility (XCF) is a z/OS component, not a piece of hardware, despite the name. Programs that want to cooperate across systems join XCF and get three kinds of service:
- Group services: a program becomes a *member* of a named *group*. A Db2 data sharing group, the JES2 multi-access spool and global resource serialisation (GRS) all use XCF groups.
- Signalling services: members send messages to each other over signalling paths, which are either channel-to-channel (CTC) connections or list structures in a coupling facility.
- Status monitoring: each system regularly writes a heartbeat into the sysplex couple data set. If a system stops updating it, the others notice the missing status update and the sysplex can remove (partition) that system.
Couple data sets
Couple data sets are special datasets on DASD that every system in the sysplex can reach. Each type holds one kind of shared information. They are formatted with the IBM utility IXCL1DSU and named in the COUPLExx member of parmlib. Every type should have a primary and an alternate on different volumes, ideally behind different storage controllers.
| Type | What it holds | Notes |
|---|---|---|
| SYSPLEX | System status, heartbeats, XCF group and member information | Required for any multi-system sysplex |
| CFRM | Coupling facility resource management policy: which CFs exist and which structures go where | Required for a Parallel Sysplex |
| SFM | Sysplex failure management policy: what to do when a system or connection fails | Strongly recommended so failures are handled automatically |
| ARM | Automatic restart manager policy | Restarts failed jobs and started tasks, optionally on another system |
| LOGR | System logger log stream definitions | Used by CICS logs, OPERLOG and others |
| WLM | The WLM service definition | One set of goals for the whole sysplex |
There is also a couple data set type for the Unix System Services shared file system. Which types a site uses depends on its configuration.
Looking at the sysplex
D XCF,SYSPLEX,ALL SYSPLEX PLEXP1 SYSTEM STATUS LAST STATUS UPDATE SYSA ACTIVE 10/08/2026 02:14:07 SYSB ACTIVE 10/08/2026 02:14:06 SYSC ACTIVE 10/08/2026 02:14:07
D XCF,SYSPLEX,ALLD XCF,COUPLED XCF,GROUPD XCF,PATHIN / D XCF,PATHOUTOne clock for everyone
Members of a sysplex write timestamps into logs that may later be merged for recovery, so all systems must agree on the time. Today this is done by Server Time Protocol (STP), a machine feature that synchronises CPCs over coupling links. It replaced the older external Sysplex Timer. In an STP network, servers take roles such as Preferred Time Server, Backup Time Server and Arbiter, and the server currently acting as the time source is the Current Time Server. D ETR shows the timing status from z/OS.
Which z/OS component (three letters) lets programs on different systems in a sysplex join groups, signal each other and monitor status?
Show a hint
Cross-system coupling facility.
Show the solution
XCF, the cross-system coupling facility, a z/OS component that provides group, signalling and status monitoring services.
Common mistakes
XCF is a z/OS component. The coupling facility is a separate LPAR. Confusing them leads to looking in the wrong place when signalling fails.
A primary with no alternate is a single point of failure for the whole sysplex. Check D XCF,COUPLE after every incident and every change.
One machine failure then takes every member down. Real availability needs members spread across machines.
What you will see at work
- Systems programmers own the COUPLExx member, the couple data sets and the policies stored in them.
- Operators run D XCF displays as a health check, especially after an IPL or a hardware change.
- Application teams meet the sysplex indirectly: their CICS regions, Db2 members and MQ queue managers run on several systems at once.
Key terms
Check your understanding.
Take this lesson's quiz and save your progress. Free.