Dataset I/O collection tuning

Excessive CPU usage is likely when both the following conditions exist:

and

You can reduce CPU usage by tuning dataset I/O collection, which allows you to get millisecond response time information at the dataset level. Use one or both of the following methods:

• Enable or disable dataset I/O collection at the volume level. In general, you should monitor only volumes for which dataset response time is a critical issue or volumes that are known to have problems.
• You can regulate dataset I/O collection by using parameters that specify when dataset level I/O monitoring starts for a volume. Apply these parameters whenever you need to monitor a large number of volumes.

You can collect dataset level I/O statistics for a device, using PARMGEN to set parameters to values that will help reduce CPU usage. The way to do this is by limiting the scope of dataset I/O monitoring.

Limiting the scope of dataset I/O monitoring

You can reduce CPU usage by limiting your dataset I/O monitoring to specific critical volumes or jobs or by setting up monitoring to trigger only when response time is poor on a particular volume.

You can limit the volumes by specifying a volser (or volser mask) for parameter KDF_FM01_VOL, or a range of addresses using KDF_FM01_FIRST_DEV and KDF_FM01_LAST_DEV. You can even specify multiple rows of definitions if your volumes do not fit in under a volser mask or device range, as in the following example:

KDF_FM            BEGIN 
KDF_FM01_ROW          01
KDF_FM01_VOL          "TSO*"
KDF_FM01_SAM_CNT      1
KDF_FM02_ROW          02
KDF_FM02_VOL          "DVP101"
KDF_FM02_SAM_CNT      1
KDF_FM03_ROW          03
KDF_FM03_VOL          "BVG288"
KDF_FM03_SAM_CNT      1
KDF_FM END

To establish a triggering response time threshold that will cause monitoring to begin on any volume that exceeds it, you must use a combination of the MSR parameter (to specify the threshold millisecond response time) and MSRTARG (the number of times a volume must exceed the MSR value in 100 consecutive samples for monitoring to be turned on). Once monitoring begins for a volume, it will continue until 100 consecutive samples are taken in which the volume does not exceed the threshold. The following example will cause monitoring to begin on a volume when its response time exceeds 20 milliseconds 51 times in 100 consecutive samples:

KDF_FM            BEGIN
KDF_FM01_ROW           01
KDF_FM01_VOL           "*"
KDF_FM01_MON_STAT      MSR
KDF_FM01_SAM_CNT       20
KDF_MSR_TRIP_CNT       51
KDF_FM                 END

To specify dataset I/O monitoring only for specific jobs, you should define these jobs in the “Application Summary” workspace in the TEP and specify "I/O Monitor Status" = "Start" in the definition dialog. This will save resources by monitoring only the datasets used by this job (or jobs if the definition uses a job name mask).