MPL Policy Example

This example shows how to use subtype 2 service class period data to analyze a first period TSO problem. The example shows how SRM used the data to resolve the problem. You can use the data in the same way to analyze why SRM isn't solving a problem. From this kind of information, a service administrator can decide whether to change goals or service class importance in the service policy.

This subtype 2 data shows the controls SRM is using for first period TSO, and the resulting performance delays.

Subtype 2: Service Class TSO Period Data

CLASS: TSO       PERIOD: 1  IMPORTANCE: 2  GOAL TYPE: SHORT RESPONSE TIME

TIME      SPI   LPI DP  MPLI MPLO SWPT  PSI EXP POL  LARGEST DELAYS
13:15:01  .10   .09 251    3    6 15952   0 S L S S  OTHR/60 ASWP/55 MPLD/2
13:15:11 6.18  6.73 251    6    9 15952   0 S L S S  ASWP/48 CPUD/44 MPLD/42
13:15:22 2.20  2.42 251    8   12 15952   0 S L S S  CPUD/38 MPLD/34 ASWP/2
13:15:32  .06   .06 251    8   12 15952   0 S L S S  OTHR/13 CPUD/3 MPLD/2
13:15:42  .06   .06 251    8   12 15952   0 S L S S  OTHR/22 MPLD/3 CPUD/2
13:15:52  .16   .16 251    8   12 15952   0 S L S S  OTHR/28 MPLD/1
13:16:03  .16   .16 251    8   12 15952   0 S L S S  OTHR/53 ASWP/6 CPUD/2

At 13:15:01 TSO period 1 was meeting its goals easily. This is indicated by a sysplex performance index (SPI) of 0.1 and a local performance index (LPI) of 0.09. The dispatching priority (DP) was 251. The MPL in and out targets (MPLI/MPLO) were 3 and 6. After being swapped out, work in the period was protected in processor storage for 15.952 seconds (SWPT=15952). There was no period wide storage isolation (PSI=0) and the expanded policy was space available for swap working set, VIO, and hiperspace pages, and LRU for demand pages (EXP POL= S L S S). At 13:15:01 TSO period 1 has a swap delay and an MPL delay but was meeting its goals easily.

Conditions change between 13:15:01 and 13:15:11. The sysplex performance index spiked to 6.18 and the local performance index was worse. This period needs help. The delay samples show that the problem could be either swap delay, processor delay, or MPL delay. Because it is the work furthest from meeting its goals, it is selected as a receiver.

The following data, from the subtype 1 record at 13:15:11 shows what happens next. The fields are explained in the first example.

Subtype 1: Trace Data Output

CLASS    P     SPI   LPI  ACTION
TSO      1    6.18  6.73   270  pa_rec_cand
TSO      1    6.18  6.73  2540  pa_prt_na_rec_val
TSO      1    6.18  6.73   850  pa_pro_na_no_donor
                           290  pa_use_disc_cent
TSO      1    4.69  5.10  3530  pa_inc_mpl

TSO period 1 is selected as the receiver candidate. The trace entry, pa_rec_cand indicates this.

The largest delay is selected to be worked on first. In this case the largest delay in recent history is swap delay. The swap delay plot is shown below. ccc indicates the current plot point. The current plot point shows that only 89 milliseconds of swap delay per transaction could be eliminated even if all swap delay were eliminated. The pa_prt_na_rec_val trace entry indicates that there was insufficient receiver value to be gained by increasing the swap protect time. The swap delay plot data is from the subtype 3 record.

Subtype 3: Swap Delay Plot

CLASS: TSO       PERIOD: 1

SWAP DELAY PLOT                   ccc
SWAP DELAY           286    209    89     76     70     24      0
TIME IN PROC STOR   3572   4021 14407  23673  40803  71248  97323

The next largest delay is processor delay. However TSO period 1 is running alone at the highest dispatching priority in use. Therefore there is no work to donate processor time. This reason for lack of action is indicated by the pa_pro_na_no_donor trace. The dispatch priorities are from the subtype 2 records.

Subtype 2: Dispatching Priority Data

CLASS    P   DP
NRBATCH  1   247
NRBATCH  2   243
NRBATCH  3   192
OFFBATCH 1   192
TSO      1   251
TSO      2   247
TSO      3   247
TSO      4   243
$SRMDISC 1   192

The third largest delay is MPL delay. The MPL delay plot below shows that here there is value to increasing the MPL. The third entry in the MPL plot, indicated by ccc shows that on average, only 48/100ths of the ready users have MPL slots. This results in an MPL delay of 202 milliseconds per completion. This plot in recorded in the subtype 3 record. The fact that MPL targets were increased is indicated by the pa_inc_mpl trace. The new MPL targets are recorded in the subtype 2 record.

MPL Delay Plot

CLASS: TSO       PERIOD: 1

MPL DELAY PLOT                 ccc
MPL DELAY            500  451  202   79    0    0    0    0    0    0
MPL SLOT PERCENTAGE    3   37   48   72   99  109  115  122  129  133

At 13:15:22 the performance index was improving but there was still significant MPL delay and the MPL targets were increased again. At 13:15:32 the work was back to meeting its goals as shown by sysplex and local performance indexes of less than 1.0.