6. DATA SOURCES › 6.3 CA MICS SMF Requirements and Considerations › 6.3.2 Interval Recording
6.3.2 Interval Recording
Interval recording is an option that is available with the
SMF type 30 records. It provides the capability to record
step information prior to the end of a step or session.
There are two major benefits realized with interval
recording:
o The amount of data lost due to a system crash is
minimized.
For example, an organization charges for a long-running
VTAM application based upon CPU usage. They specify a
recording interval of 30 minutes, which causes an
interval record to be written to SMF every 30 minutes.
The system is typically started Monday morning and
brought down Friday night. The system crashes Friday
after lunch. Without interval accounting, an entire
week's worth of charges are lost, since the long-running
application's step end record is not written to the SMF
data sets due to the crash. With interval accounting, a
maximum of 29.99 minutes of charges is lost.
o The resource utilization of long running address spaces
is sliced into many discrete records. This provides a
better determination of exactly when the resources were
consumed.
For example, a six-hour step completes execution, but
capacity planning requires information on when the
resources were actually consumed. If interval recording
of 30 minutes is specified for batch jobs, a record with
resource utilization measures for each 30-minute
interval will be written to the SMF data sets.
There are two types of SMF type 30 interval records:
SMF type 30 subtype 2 - Complete Interval
SMF type 30 subtype 3 - Last/Partial Interval
Interval recording may be specified with different time
limits for each subsystem (Batch, TSO, Started Tasks,
APPC/MVS Transaction Programs (TPs)) in the SMFPRMxx member
of SYS1.PARMLIB (see Section 6.2.1.3). Most data centers
use intervals of anywhere from 10 minutes to one hour,
depending on the importance and use of the data.
CA MICS will use interval type 30 records (subtypes 2 and 3)
instead of step end type 30 records (subtype 4) to build the
BATPGM, BAT_TS, BAT_ST, and BAT_TP files in the CA MICS
database if both types are present. This will expand the
size of the DETAIL timespan files since multiple CA MICS
records will be created for steps and sessions that run
longer than the interval specified.
SYNCHRONIZING SMF INTERVAL RECORDING
SMF can be instructed to force individual address space
intervals to end at specific, synchronized points in time
rather than at times associated with the start of each
individual address space.
This feature allows better analysis of system-wide activity
at the address-space level. With interval record
synchronization, the start and end time of interval records
can be forced to fall neatly between hourly boundaries. This
eliminates the problem of how to distribute the resources of
an interval that spans an hour. For example, if an interval
began at 9:45 and ended at 10:15, the best you could do was
to assume that half of the measurements occurred in hour 9
and the other half in hour 10.
One consequence of SMF interval record synchronization is
that the first interval record for an address space will not
represent a complete interval. The first record is instead
produced when the SYNCH time is reached. At that time,
intervals end for all address spaces. The second through N-1
records will represent complete intervals. The final record
will again be less than a complete interval unless the step
ends precisely at the interval expiration time. This is
illustrated in the following example.
Step Step
Start End
INT-1 INT-2 INT-3 INT-4
|<---->|<-------->|<-------->|<-->|
8:13 8:29 8:59 9:29 9:41
In the example, both intervals 1 and 4 are less than the 30
minute length of complete intervals.
Another consequence of SMF interval record synchronization is
that it can cause gaps in the interval record history of an
address space. When interval synchronization is in effect,
interval expiration is not a valid reason to swap in an
inactive address space. Instead, interval-end activities and
interval record writing occur when the address space is
swapped in due to satisfaction of the condition that
originally caused the swap out.
When an address space remains continuously swapped out for
one or more intervals, SMF does not write records for those
intervals. This leaves gaps in the SMF interval record
history for the address space. The following example
illustrates how this occurs.
SMF
Interval Interval Buffer
Start End Write
|<---------->|--------------|
8:29 8:59 9:33
Address Space Continually Swapped Out
|<--------------->|
8:52 9:33
Interval Interval
Start End
|<---------->|
9:29 9:59
In the example, the interval time is 30 minutes, and the
SYNCH time is specified at 59 minutes after the hour. This
means that intervals end at 29 and 59 minutes after every
hour. During the 8:29 to 8:59 interval, the address space
was swapped out (at 8:52). It remained swapped out for 41
minutes and was swapped in at 9:33. At 9:33, an SMF record
is written for the 8:29-8:59 interval. Statistics begin
collecting for the 9:29-9:59 interval, but no record is
produced for the 8:59-9:29 interval.
CA MICS detects these gaps and records the total gap time in
data element Interval Skipped Time (PGMSKPTM). This value is
recorded in the first interval record following the gap. In
the example above, the CA MICS database record for the
9:29-9:59 interval would contain PGMSKPTM=1800 (1800
seconds=30 minutes).
Finally, when synchronization is active, you should not rely
on the date and time that the interval record is written to
the SMF buffer as a measure of interval expiration date and
time. The SMF type 30 interval records contain the actual
interval end timestamp. These fields are used for the
STARTTS and ENDTS of CA MICS step-level observations created
from interval records.
