The impact analysis display format

The impact analysis display has four parts.

Figure 1. Impact analysis display format

1 [IANL PAY0022                                 IAST=2.2 IATH=5 IACS=1 IACL=5
2 [Productive/Contention/Other   | 10.8% 83.7% 5.5%  |      | 38.7% 52.5% 8.8%
3 [Contending Address Spaces     |Short (3:24M,88)   |      |Long (13:24M,361) 
  @------------------------------+-1-2-3-4-5-6-7-8-9-+      {-1-2-3-4-5-6-7-8-9-+
  |  *Total Contention*    *83.7*|------======>>>>>  |*52.2*|------====>        |
  |  *Self-Contention*       10.2|-> . . . . . . . . |   7.9|-> . . . . . . . . | 
4 |  WORK0033 (J1208)        37.5|-----=>. . . . . . |   0.0| . . . . . . . . . |
  |  TESTCICS (S1366)        13.3|-->. . . . . . . . |  11.0|-> . . . . . . . . |
  |  USER01   (T0934)         0.0| . . . . . . . . . |  15.3|-->. . . . . . . . |
  %  System                  22.7|---->. . . . . . . |  18.3|---> . . . . . . . |
  1. Input command and options

    In the example shown in Figure 1, the input command is IANL PAY0022, which requests an overview display for workload PAY0022. The remainder of the line contains information about the collection and display parameters in effect. For this command, the parameters are set as follows:

    • The sampling interval (IAST) is 2.2 seconds.
    • The plot threshold (IATH) is 5%.
    • The short term interval (IACS) is 1 minute.
    • The long term interval (IACL) is 5 short-term intervals (about 5 minutes).

    These parameters and their default settings are discussed in Setting defaults.

    The next two lines, the productivity analysis line and the interval description line, should be read as a pair. When read together, these two lines give you a general idea of how reliable your data is, and how the workload (or group) is being impacted by other address spaces in the z/OS® system. Long-term and short-term contention are shown side-by-side; long-term figures are on the right side of the display under the Long heading, and short-term figures are in the center of the display under the Short heading.

  2. The productivity analysis line

    Impact analysis analyzes workload productivity by sampling the execution state of the monitored workload at periodic intervals. By taking a large number of samples, it is possible to draw valid inferences about how a workload is spending its time.

    The productivity analysis line of the impact analysis display provides a breakdown of how the impacted workload has been spending its time, in both the long and short term.

    Three categories of time are represented on impact analysis displays:
    Productive
    Time spent doing desirable activities such as using CPU or active I/O. In this example, the workload has been spending 10.8% of its time in the short term doing productive work, and 38.7% in the long term.
    Contention
    Time spent waiting for other resources that are being used by other z/OS address spaces. This type of wait is referred to as a chargeable wait, and is associated with the following execution states:

    • CPU waits (waiting for dispatch by z/OS)
    • Queued I/O
    • Reserves (charged only to system)
    • Tape or disk mount pending
    • Private area (local) page waits
    • Common area page waits
    • Waits for swap page-ins
    • Enqueue waits
    • SRM delays
    • Swaps due to real storage shortages
    • Cross-memory local lock waits (CML)

    In this example, the workload has been spending 83.5% of its time in the short term waiting for contending address spaces, and 52.5% in the long term.

    Other
    Time during which the workload is waiting, but the wait time is not attributable to another address space. For example, a tight CPU loop unnecessarily executes instructions, wasting time that would be put to better use processing data. Although the address space stuck in a tight loop is degraded, no other address space is responsible.

    This type of wait is referred to as a nonchargeable wait, and is associated with the following execution states:

    • Waiting for MSS staging
    • Local or global lock waits (except CML)
    • SRM delays (RTO)
    • Swaps due to terminal output waits
    • Swaps due to detected waits
    • Requested swaps
    • Transition swaps

    In the example, this type of wait time accounts for 5.5% of the short term and 8.8% of the long term.

    Another type of time that is associated with productivity, but is not reported on impact analysis displays, is referred to as idle time. Idle time is time in which the monitored workload is not doing work, but is not being prevented from doing so. Idle time is associated with the following execution states:

    • ECB waits
    • ECB waits with STIMER
    • STIMER waits
    • Long wait swaps
    • HSM waits
    • JES2 waits

    By looking at the productivity analysis in the long and short term, you can understand how workload productivity is changing. In this example, contention has been more severe in the short term than in the long term, accounting for 83.7% of the non-idle time, as opposed to 52.5% in the long term. Similarly, productive time has dropped from 38.7% in the long term to 10.8% in the short term.

  3. The interval description line

    The interval description line, which immediately follows the productivity analysis line, shows the length of the short and long term intervals, and the number of observations in each interval.

    The short term interval side of the display shows the length (in minutes) of the short term interval, and the number next to it indicates the number of observations made during that period.

    The long term interval side of the display shows the length (in minutes) of the long term interval, and the number next to it indicates the number of observations made during that period.

    In this example, the short term interval contains information collected over the past 3 minutes and 24 seconds, during which there were 88 observations. Similarly, the long term interval spans 13 minutes and 24 seconds, during which 361 observations were made.

    Since the reliability of the data improves with the number of observations, it is helpful to note the number of observations in the intervals. In general, you should wait until the display shows at least 30 observations per interval before relying on this data.

  4. Display body

    The body of the display shows various contenders and the degree to which they are impacting the workload. The number next to each impact source indicates the percent of the total non-idle time that the workload was found waiting for one or more resources while that contender was using it.

    The arrows to the right of the numbers provide graphic representations of the data in the following ways:
    • If the contention is less than thirty percent, the arrow is made up of hyphens only.
    • If contention is greater than thirty percent, the arrow between thirty and sixty percent is made up of equal signs.
    • Contention greater than sixty percent is designated by greater-than (>) signs.

    On extended color terminals, the arrows are displayed in color. The colors for the three contention levels are green, yellow, and red, respectively. For information about turning on extended color, see The User Profile Facility.

    The first line of this section shows total contention. (Total contention is also shown as part of the productivity analysis line. It is provided here for easy comparison with other lines in the display body.) The total contention shown on this line may not be equal to the sum of the contention from all impact sources shown underneath it. This is because only the main impact sources are shown on the default display (only those that account for 5% or more of the non-idle time are included). You can adjust this threshold with the IATH command, as described in Setting defaults.

    The next three lines of the sample display show address spaces that are competing with PAY0022 for resources. The display lists competing address spaces in order of decreasing short term impact. The most important contenders are shown first.

    Below the short-term contenders, the display lists long-term contenders that are not causing any short term contention. Long-term contenders are also listed in order of decreasing impact.

    The example shows three impact sources: a batch job (WORK0033), a started task (TESTCICS), and a TSO user (USER01). Next to each name, the display shows the JES2 job number of the impact source. The job numbers each begin with a letter that denotes the type of job:

    J
    identifies a batch job
    S
    indicates a started task
    T
    identifies a TSO user

    In this example, WORK0033 is the largest short term impact source, accounting for 37.5 percent of non-idle time. However, the long term display shows USER01 to be the major impact source over a longer period of time, even though it is not currently impacting the monitored workload.

    The last line of the display shows system contention. This is primarily due to contention that is detected but cannot be attributed to any particular address space. You can minimize the system contention value by running OMEGAMON at a high dispatching priority. The following section provides more information on system contention.