Previous Topic: 5.4.7.2 Data Elements List

Next Topic: 5.4.7.4 Retrieval Examples

5.4.7.3 Usage Considerations


Special considerations or techniques related to using the
WLMSEC file are provided below.

1.  Care must be exercised in using the special date and time
    data elements contained in each CA MICS file. As the
    file's granularity increases in higher timespans, certain
    fields lose significance; they should not be used in
    those cases.

    o  HOUR should not be used in MONTHS.
    o  DAY and DAYNAME should not be used in WEEKS, MONTHS.
    o  WEEK should not be used in MONTHS.

2.  ENDTS and STARTTS, when appearing in the DAYS, WEEKS, or
    MONTHS timespans, bound the span of time over which the
    data has been summarized.  STARTTS is the lowest date and
    time, and ENDTS is the highest date and time for the data
    summarized.  These data elements have different meanings
    when used in the DETAIL timespan versus the DAYS, WEEKS,
    and MONTHS timespans.  Their purpose in DETAIL is
    described below:

    o  STARTTS represents RMF recording interval start time.
    o  ENDTS represents RMF recording interval end time.

3.  z/OS CPU Time Considerations Related to WLM Analysis

    An important aspect of performance management is
    capturing performance data for report generation.  z/OS
    collects CPU time consumed by dispatchable units such as
    tasks (TCBs) and service requests (SRBs) associated to
    executed transactions.

    Below is a diagram of the CPU time components and how
    they relate to the various CPU time related data elements
    in the WLMSEC file (this diagram also applies to report
    classes in the WLM_EC file):

ELAPSED TIME !----------------------------------------(SECTOTET)-------------------------------------------------------------! CPU TIME !-----------------------------------------------------------------------------------------! CAPTURED IN SERVICE UNITS <-------------(SECTCPTM)-------------------> PREEMPT NON PREEMPT WAIT (TCB) (SRB) RCT IIT HST !--------!----CP---+--ZAAP--+--ZIIP---!-----------!----------!----------!----------!-------------!------------! (SECCPUTM, SECZAPTM, SECSUPTM) (SECSRBTM) (SECTORCT) (SECTOIIT) (SECTOHST) <--------------------------------><------------><-----------> CAPTURED NOT IN UNCAPTURED LPAR STOP SERVICE UNITS



    When a TCB or an SRB is dispatched, an x value is loaded
    in the CPU timer (a hardware clock that decreases its
    contents).  When the TCB or SRB is interrupted, the
    current CPU timer (x-y) is stored in memory.  Then, y is
    accumulated in the ASCB control block (or in an enclave
    ENCB control block).  There are two counters in ASCB: one
    for preemptable dispatchable units, such as TCBs, SRB
    clients, SRB enclaves, and one for non-preemptable
    dispatchable units, such as traditional SRBs.

    The accumulated time is called captured time and is
    converted into CPU service units.  It typically accounts
    for 85-90% of the total CPU time for systems with a high
    utilization.  The value can be significantly smaller in
    under-utilized systems depending on the activation of the
    Alternate Wait Management.

    z/OS (or the LPAR) also accounts for the CP wait time
    (logical and physical), when the CPU is placed in wait
    state because there is no work to be done.  Then, wait
    time is the time that the logical CP was in wait state
    due to the lack of activity in the logical partition.

    There is also a captured time not converted in CPU SUs.
    It is the captured CPU time not included in ASCB
    preemptable or non-preemptable counters, and consequently
    not converted to CPU SUs.  The three components
    generating such times are the following:

    o  IIT, the CPU time spent in I/O SLIH routine

    o  RCT, the CPU time spent in the address space region
       control task

    o  HST, the CPU time spent accessing hyperspace

    There are also other activities executed in the CPU that
    are not accounted by z/OS.  This time is called
    non-captured time.  It can be numerically derived by
    subtracting the captured time from the total busy time.
    Non captured time includes the following:

    o  Overhead such as page faults and swaps

    o  Global services such as dispatcher, WLM sampling, SRM,
       where there is no address space or enclave in charge

    o  Services where it is impossible to find the task that
       requests the service, such as I/O FLIH processing

    o  LPAR processing associated with the logical partition