Checking Page Fault Rates (iSeries)

If data is already in main storage, it can be accessed independently of the main storage pool in which it is stored. However, if required data is not available in a main storage pool, it is brought into the same main storage pool as the job that referenced it. (This is called a page fault.) When data is transferred from auxiliary storage into a memory pool, other data is replaced and - provided the data was changed - stored in auxiliary storage. (This is called paging.)

On iSeries [Pages in] = [Pages out].

Since page faults adversely affect performance, you must make sure that the main storage pool is large enough to keep paging to an acceptable level.

Value limits for page faults on iSeries

If you notice that there is a high page fault rate in a main storage pool, increase the size of the pool. (To see information about page faults, use the operating system monitor (Transaction ST06) or alert monitor (Transaction RZ20).

Pool	Value Limit
Machine pool (always system pool 1)	Since the machine pool contains objects used throughout the whole system, page faults in the machine pool affect all the jobs in the system. For this reason, make sure that the page fault rate is as low as possible in this pool. Upper limit: 10 page faults per second
Other pools	For optimum performance, an average rate of 2 page faults per second for each active work process should not be exceeded. The upper limit that should not be exceeded is a rate of 20 page faults per second for each active work process. (see example below)

The following examples can help you when determining page fault rates in your system:

Example for Best Performance

Page faults per second in the memory pool	20
Active work processes (that is, iSeries jobs with status RUN) in the memory pool	10
Disk service time	10 ms (average disk service time). Substitute the current value from your system for this value.
Result	2 page faults/s * 10 ms = 20 ms/s paging

This corresponds to a paging overhead of approximately 2%.

Example for Upper Limit

Page faults per second in the memory pool	200
Active work processes (that is, iSeries jobs with status RUN) in the memory pool	10
Disk service time	10 ms (average disk service time). Substitute the current value from your system for this value.
Result	20 page faults/s * 10 ms = 200 ms/s paging

This corresponds to a paging overhead of approximately 20%. The 20% limit should not be exceeded.