The measurements that matter
Each layer has a few measurements that tell you most of what you need: for z/OS, CPU, workload goals, storage and contention; for CICS, response time, task limits and storage; for Db2, threads, locks and buffer pools; for MQ, queue depth and channel status. Knowing what normal looks like is what makes them useful.
z/OS
| Measurement | What it tells you |
|---|---|
| CPU busy (LPAR and by service class) | How much processor is in use and by whom. High CPU on z/OS is normal while goals are met. |
| WLM performance index (PI) | Whether each service class meets its goal: below 1 is better than the goal, above 1 is missing it. |
| Delay reasons | What work is waiting for: processor, I/O, storage, an enqueue, a lock. |
| Paging and auxiliary storage | Whether real storage is short; D ASM shows how full the page data sets are. |
| Enqueue contention | Jobs or users waiting for a data set or resource another holds. |
| Spool use | How full JES2's spool is; a full spool stops work. |
| I/O response time | How long disk requests take, by device or storage group. |
CICS
- Transaction rate and response time by transaction: the users' experience.
- Tasks against the maximum (MXT): at the limit, new transactions queue.
- Short-on-storage conditions: CICS is running out of a storage area.
- Abend counts by transaction, such as a rise in ASRA.
Db2
- Threads in use against the maximum, by connection (CICS, batch, distributed).
- Lock suspensions, timeouts and deadlocks: contention between work, often batch against online.
- Buffer pool activity: getpages, synchronous reads and hit ratios; a sudden change can mean a new access path.
- Log activity and archive status.
MQ
- Queue depth (CURDEPTH) against the maximum (MAXDEPTH), and how old the oldest message is.
- Channel status: RUNNING is healthy; RETRYING or STOPPED means messages are not moving between queue managers.
- Dead-letter queue depth: messages that could not be delivered.
A service class has a performance index of 1.8. Is it meeting its goal? (yes or no)
Show a hint
Above 1 means...
Show the solution
No: above 1 means it is missing its goal.
Common mistakes
z/OS is designed to run busy. Check whether important work is meeting its goals.
An average response time can look fine while one transaction is very slow. Look at the worst transactions too.
A filling queue usually means its consumer has stopped. Find the consumer.
What you will see at work
- Teams keep a short list of 'vital signs' per system, with normal ranges for each time of day.
- Many incidents start as contention: a batch job running into the online day and holding Db2 locks is a classic example.
- Capacity planners use the same measurements over months to decide when to add capacity.
Key terms
Check your understanding.
Take this lesson's quiz and save your progress. Free.