WLM in production
WLM problems show up as slow transactions or late batch. SDSF shows each job's service class, RMF reports show whether classes meet their goals, and a few commands display and change what WLM is doing.
Where to look
| Tool | What it shows |
|---|---|
| SDSF DA panel | Running work with its service class and current resource use |
| RMF Monitor III | Near real-time delays: what each job or class is waiting for |
| RMF Workload Activity report | Goal, achieved value and PI per service class period over an interval |
| SMF type 72 records | The WLM workload data RMF reports are built from, kept for history and capacity planning |
Commands
| Command | What it does |
|---|---|
| D WLM | Displays the active service policy and WLM status |
| V WLM,POLICY=name | Activates a different service policy (change-controlled) |
| E jobname,SRVCLASS=class | RESET command: moves a running job or task to another service class |
| E jobname,QUIESCE | Quiesces a job: it gets resources only when nothing else wants them |
| E jobname,RESUME | Returns a quiesced or reset job to its classified service class |
E is the short form of the RESET command. SDSF can also change a job's service class by overtyping it, if you are authorised. These changes last only for that job's run; the classification rules are unchanged.
zIIP processors
A zIIP is a specialty processor that runs certain eligible work, such as parts of DB2 distributed and utility processing, Java and some system work, at lower software cost than general processors. WLM dispatches eligible work to zIIPs. When zIIPs are busy, eligible work may run on general processors depending on configuration, which shows up in RMF reports.
Scenario: month-end slowdown
Capping, briefly
Sites can limit processor use with resource groups (a cap on a set of service classes) and with defined capacity or group capacity limits on LPARs, often to control software costs. When work is capped it shows as delay in RMF even though the machine has spare capacity: worth checking when 'the system looks idle but my work is slow'.
Common mistakes
A slow job in the wrong service class looks exactly like a slow program. Check SRVCLASS in SDSF early.
RESET lasts for one run. Next month the same job lands in the same wrong class unless the rule is fixed.
If work is capped by a resource group or LPAR limit, adding more importance does nothing. Look at capping delays in RMF.
What you will see at work
- Performance analysts review RMF workload reports daily or weekly; production support uses SDSF and Monitor III during incidents.
- Capacity planners use SMF 72 history to forecast when goals will stop being met.
- Many incident reviews end with an action to fix a classification rule or a naming standard.
Key terms
Check your understanding.
Take this lesson's quiz and save your progress. Free.