Author: Roy Boxwell

OK, I kept you all waiting long enough… Here are my AI results with Db2 13 FL501!

Start at the Start

We begin with the beginning as last time:

Let’s get Connected!

Here you can see that I have already defined my little test Db2 13 system to the system:

Join the Dots …

Now just click on the vertical dots:

Here you can Disconnect, Edit (Which shows you the same window as “add connection”), List AI objects or Delete.

What do we have?

Choosing List AI objects you see what has been created:

Clicking on the down arrow on the left-hand side to expand looks a lot better than last month:

Clickedy-click-click

Now, clicking on the vertical dots on the right hand side, you can choose to Disable AI query or Enable AI query. (I have actually added a new column for consideration, so first I clicked on Disable and then clicked again on Enable)

Just the Facts, Ma’am – Again

Here you must make your “Usual Suspects” decision: which columns to actually use in building the AI Model. I am using our WorkLoadExpert performance table in this newsletter and have selected 17 columns that I think will work together nicely. Only one can be a “Key” column – I choose STMT_ID in this case. Once you are done selecting columns, click on the big blue “Next” button where you may then add additional filters to remove any rows you know are to be ignored:

Playing Chicken?

When done, click on the big blue “Enable” button and you get your last chance to chicken out:

SIO and CPU Records!

Click here and then get a cup of coffee….or go to SDSF and marvel at how much CPU and IO Spark actually uses and does this as the light bulbs dim in your part of the world…

You Keep me Spinning

Oddly, at least when I do this, the Initializing spinning wheels:

Right Round and Around

… never stop. The WLM Stored procedure for utilities was finally kicked off about 40 minutes later:

                    J E S 2  J O B  L O G  --  S Y S T E M  
                                                             
10.43.19 STC09611 ---- WEDNESDAY, 01 MAR 2023 ----           
10.43.19 STC09611  $HASP373 DD10WLMU STARTED                 
10.43.19 STC09611  IEF403I DD10WLMU - STARTED - TIME=10.43.19
10.43.19 STC09611  ICH70001I SQLDIID  LAST ACCESS AT 09:37:37

A Loaded Question?

And loaded all the required data:

ICE134I 0 NUMBER OF BYTES SORTED: 99083595                  
ICE253I 0 RECORDS SORTED - PROCESSED: 49173, EXPECTED: 49173

A quick exit and re-logon to the web interface…and Tra la!

Not only AI but Dr Who!

It is also strange that it seems to be in a time machine, one hour in advance of my local…Anyways, my new data is there and so onward! (I have since heard that our time zone setting is actually to blame and that just going back one level, and then forward again, stops the spinning wheel problem. However, just wait until Spark finishes and the stored procedure has loaded your data!)

Never Trust a Statistic You haven’t Faked Yourself!

Clicking on Data statistics shows:

Influencer of the Day?

Then you can look at the Column influence:

Super Model?

Back at the top you can then review the Model details:

Or just a Cluster….

Here are the Cluster center details:

Going back to the List AI Objects window, there are two blue buttons: Add object and Run query. I did not discuss Run Query last month but it gives you a SPUFI-like ability on the PC, tailored to the AI BiFs:

Lets RUN Away!

Clicking on Query type gives a drop-down list of the basic AI BiFs where it then gives you an example SQL (based on the documentation, *not* on any AI Tables you might have done!). Once you type in any query the “run” box turns blue:

It Works!

Click run and see the results:

Data Review

Once the model is trained, you can then review on the host what it has done. In SPUFI you can find details of what you have done in the pseudo Db2 catalog tables that support Data Insights, (I have removed a ton of rows to make this readable – sort of!):

SELECT * FROM 
SYSAIDB.SYSAIOBJECTS ; 
---------+---------+---------+---------+---------+---------+---------+---------+-------+-
              OBJECT_ID  OBJECT_NAME                       OBJECT_TYPE  SCHEMA  NAME     
---------+---------+---------+---------+---------+---------+---------+---------+-------+-
                     26  --------------------------------  T            IQA061QB IQATW001

-------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+
STATUS                   CONFIGURATION_ID                 MODEL_ID  CREATED_BY                        CREATED_DATE              
-------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+
Enabled                                36                       36  SQLDIID                           2023-02-24-07.57.42.086932

-------+---------+---------+---------+---------+---------+---------+---------+
LAST_UPDATED_BY                   LAST_UPDATED_DATE           DESCRIPTION     
-------+---------+---------+---------+---------+---------+---------+---------+
SQLDIID                           2023-03-01-10.43.38.407460  ----------------



SELECT * FROM  
SYSAIDB.SYSAICONFIGURATIONS ; 
---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+-----
       CONFIGURATION_ID  NAME                                            OBJECT_ID  RETRAIN_INTERVAL  KEEP_ROWIDENTIFIER_KEY 
---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+-----
                     36  --------------------------------                       26  ----------------  Y                      
---+---------+-------
NEGLECT_VALUES       
---+---------+-------
                     
---+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+--------
CREATED_BY                        CREATED_DATE                LAST_UPDATED_BY                   LAST_UPDATED_DATE         
---+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+--------
SQLDIID                           2023-03-01-09.51.00.994421  SQLDIID                           2023-03-01-09.51.00.994461



SELECT * FROM 
SYSAIDB.SYSAICOLUMNCONFIG
ORDER BY 1 , 3 , 2 ;   

--+---------+---------+---------+---------+---------+---------+---------+---------+---------+
CONFIGURATION_ID  COLUMN_NAME              COLUMN_AISQL_TYPE  COLUMN_PRIORITY  NEGLECT_VALUES
--+---------+---------+---------+---------+---------+---------+---------+---------+---------+
              36  END_USERID               C                  H                              
              36  PRIM_AUTHOR              C                  H                              
              36  PROGRAM                  C                  H                              
              36  REF_TABLE                C                  H                              
              36  REF_TAB_QUAL             C                  H                              
              36  STMT_ORIGIN              C                  H                              
              36  STMT_TEXT                C                  H                              
              36  TRANSACTION              C                  H                              
              36  WORKSTATION              C                  H                              
              36  COPIES_NO                I                  H 
.
.
.                             
              36  WLX_TYPE                 I                  H                              
              36  WORKSTATION_OLD          I                  H                              
              36  STMT_ID                  K                  H                              
              36  CPU_TIME                 N                  H                              
              36  ELAPSE_TIME              N                  H                              
              36  EXECUTIONS               N                  H                              
              36  GETP_OPERATIONS          N                  H                              
              36  ROWS_EXAMINED            N                  H                              
              36  ROWS_PROCESSED           N                  H                              
              36  STMT_LENGTH              N                  H                              

When the column COLUMN_AISQL_TYPE has a value of “I” it means it is ignored by AI processing. Also note that this table SYSAICOLUMNCONFIG gets two extra columns (COLUMN_VECTOR_CARDINALITY and MAX_DATA_VALUE_LEN) once you apply the vector prefetch upgrade APARs:

1. For IBM Z AI Optimization (zAIO) library and IBM Z AI Embedded (zADE) library in the IBM Z Deep Neural Network (zDNN) stack on z/OS:
   • Apply OA63950 and OA63952 for z/OS 2.5 (HZAI250).
   • Apply OA63949 and OA63951 for z/OS 2.4 (HBB77C0).
2. For OpenBLAS on z/OS:
   • Apply PH49807 and PH50872 for both z/OS 2.5 and z/OS 2.4 (HTV77C0).
   • Apply PH50881 for z/OS 2.5 (HLE77D0).
   • Apply PH50880 for z/OS 2.4 (HLE77C0).
3. For Db2 13 for z/OS, apply PH51892. Follow the instructions for DDL migration outlined in the ++ HOLD text. By default, the new Db2 subsystem parameter MXAIDTCACH is set to 0, indicating that vector prefetch is disabled. To enable vector prefetch, set MXAIDTCACH to a value between 1 and 512. This parameter is online changeable. See “IBM Db2 13 for z/OS documentation” on MXAIDTCACH.
4. For SQL Data Insights 1.1.0 UI and model training (HDBDD18), apply PH51052.

Further, the table SYSAIMODELS got a new column MODEL_CODE_LEVEL and an increase in size for the METRIC column to 500K with the above APARs.

SELECT * FROM  
SYSAIDB.SYSAIMODELS ; 
---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------
               MODEL_ID  NAME                                            OBJECT_ID         CONFIGURATION_ID  VECTOR_TABLE_CREATOR
---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------
                     36  --------------------------------                       26                       36  DSNAIDB             

+---------+---------+--
VECTOR_TABLE_NAME      
+---------+---------+--
AIDB_IQA061QB_IQATW001

+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+
VECTOR_TABLE_STATUS  VECTOR_TABLE_DBID  VECTOR_TABLE_OBID  VECTOR_TABLE_IXDBID  VECTOR_TABLE_IXOBID  VECTOR_TABLE_VERSION
+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+
A                                  329                  3                  329                    4                     1

-------+---------+---------+---------+---------+---------+---------+---------+-------
METRICS                                                                              
-------+---------+---------+---------+---------+---------+---------+---------+-------
[{"discriminator":8.59443984950101,"influence":0.9367419701380996,"name":"TRANSACTION",

-------+---------+---------+
INTERPRETABILITY_STRUCT     
-------+---------+---------+

-------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+----
CREATED_BY                        CREATED_DATE                LAST_UPDATED_BY                   LAST_UPDATED_DATE         
-------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+----
SQLDIID                           2023-03-01-09.51.03.777504  SQLDIID                           2023-03-01-10.43.37.796847

---+---------+---------+---------+---------+--
MODEL_ROWID                                   
---+---------+---------+---------+---------+--
2495E518C773B081E09C018000000100000000002213  



SELECT * FROM                            
SYSAIDB.SYSAICOLUMNCENTERS 
ORDER BY 1 , 2 , 3 ;
----+---------+---------+---------+---------+---------+---------+
MODEL_ID  COLUMN_NAME                          CLUSTER_MIN  LABEL
----+---------+---------+---------+---------+---------+---------+
      36  CPU_TIME                 -0.7200000000000000E+76  EMPTY
      36  CPU_TIME                 +0.0               E+00  c0   
      36  CPU_TIME                 +0.2000000000000000E+01  c1   
      36  CPU_TIME                 +0.1617671400000000E+08  c9   
      36  ELAPSE_TIME              -0.7200000000000000E+76  EMPTY
      36  ELAPSE_TIME              +0.0               E+00  c0   
      36  ELAPSE_TIME              +0.2000000000000000E+01  c1   
      36  ELAPSE_TIME              +0.1008466600000000E+08  c9   
      36  ELAPSE_TIME              +0.1074954980000000E+09  c10  

SELECT * FROM  
SYSAIDB.SYSAITRAININGJOBS ; 
---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+--------+-------
        TRAINING_JOB_ID                OBJECT_ID         CONFIGURATION_ID                 MODEL_ID  STATUS  PROGRESS RESOURCE 
---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+--------+-------
                     33                       26                       33                       33  F              0 
                     34                       26                       34                       34  F              0 
                     35                       26                       35                       35  C            100 
                     36                       26                       36                       36  C            100 


-+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+------
MESSAGES                                                                                                                        
-+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+------
{"messages":"failed to train model: Something went wrong with the zLoad, please check the SQL DI log for more details.","resumeI
{"messages":"failed to train model: Something went wrong with the zLoad, please check the SQL DI log for more details.","resumeI
{"messages":"model training is completed","sparkSubmitId":"driver-20230224105851-0002"}                                         
{"messages":"model training is completed","sparkSubmitId":"driver-20230301085133-0003"}                                         

-+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+----
START_TIME                  END_TIME                    CREATED_BY                        CREATED_DATE              
-+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+----
2023-02-24-08.01.20.737455  2023-02-24-08.51.56.386011  SQLDIID                           2023-02-24-08.01.20.737455
2023-02-24-10.52.27.687965  2023-02-24-11.43.22.095144  SQLDIID                           2023-02-24-10.52.27.687965
2023-02-24-11.58.20.109571  2023-02-24-12.49.20.660143  SQLDIID                           2023-02-24-11.58.20.109571
2023-03-01-09.51.03.777662  2023-03-01-10.43.38.407414  SQLDIID                           2023-03-01-09.51.03.777662

---+---------+---------+---------+---------+---------+------
LAST_UPDATED_BY                   LAST_UPDATED_DATE         
---+---------+---------+---------+---------+---------+------
SQLDIID                           2023-02-24-08.51.56.386030
SQLDIID                           2023-02-24-11.43.22.095164
SQLDIID                           2023-02-24-12.49.20.660160
SQLDIID                           2023-03-01-10.43.38.407425

KPIs from my Data

Here are a few KPIs from these first test runs:

SELECT COUNT(*) FROM IQA061QB.IQATW001 ;
      64390                                                 

SELECT COUNT(*) FROM DSNAIDB.AIDB_IQA061QB_IQATW001 ;
      49173                                                 

SELECT  SUBSTR(A.COLUMN_NAME, 1, 12) AS COLUMN_NAME 
      , SUBSTR(A.VALUE      , 1, 12) AS VALUE 
      , A.VECTOR 
FROM DSNAIDB.AIDB_IQA061QB_IQATW001  A 
ORDER BY 1 , 2 ; 

---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+--
COLUMN_NAME   VALUE         VECTOR                                                                                                  
---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+--
CPU_TIME      c0            3E594822BC9D2C7A3CD4F61DBD37E5033D34B314BD4CF8E3BD4B4D47BCB6CE293D1DBA1A3D858FDF3DC4DF08BD9E77753CCED43F
CPU_TIME      c1            3D9214383CFE4C90BDB3DFE4BBE407563BBA69553DB48FEFBCF39451BC6BABF0BDA31BDFBDB52F883C30B992BC8D71AF3D9E54FF
ELAPSE_TIME   c0            3E55B744BCCC5CED3D129B14BC9E553C3C9B121EBD8949C0BD4F838DBD1582A33D36D6363DA1F72F3DBCB033BDAFB88F3D4DE348
ELAPSE_TIME   c1            3DE390AC3D2DCC98BD2DF437BC5B7F713D766D103BD1AC10BB48E2C43B9FA9E6BD80D5D7BDC40AFE3CE586C9BCACADE93DFE2745
END_USERID    BOXWEL2       3D505075BD80E40F3D3AAB60BBA463F6BBCC51C43D92B118BD044D20BD8C6B3B3CC315133BBB087A3DC1D5923DC4EB763D039C8B
END_USERID    BOXWEL3       3D2FB919BC5013E3BD6652DDBD4654DA3DA4AC83BA70024FBD7FAFD0BCF16670BB2CCB4B3DBE32E93DFE13383CB052283C82FD46

As I mentioned last month the vector tables are very “special”!

What now?

So now we have analyzed a bunch of SQL WorkLoadExpert data from our own labs. What can we do?

First up, I wish to see what user KKKKKKK does with dynamic SQL that is “similar” to what I do with table IQATW001 but I am only interested in those SQLs where the AI thinks it is more than 0.5 (so very analogous):

SELECT AI_ANALOGY('BOXWEL3'  USING MODEL COLUMN PRIM_AUTHOR,  
                  'IQATW001' USING MODEL COLUMN REF_TABLE  ,  
                  'KKKKKKK'  USING MODEL COLUMN PRIM_AUTHOR,  
                  REF_TABLE ) AS AI_VALUE                     
      ,A.WLX_TIMESTAMP                                        
      ,A.STMT_ID                                              
      ,A.STMT_TIMESTAMP                                       
      ,SUBSTR(A.PRIM_AUTHOR , 1 , 8 ) AS PRIM_AUTHOR          
      ,SUBSTR(A.PROGRAM     , 1 , 8 ) AS PROGRAM              
      ,SUBSTR(A.REF_TABLE   , 1 , 18) AS REF_TABLE            
      ,A.EXECUTIONS                                           
      ,A.GETP_OPERATIONS                                      
      ,A.ELAPSE_TIME                                          
      ,A.CPU_TIME                                             
      ,A.STMT_TEXT                                            
FROM IQA061QB.IQATW001 A                                      
WHERE A.PRIM_AUTHOR = 'KKKKKKK'                               
  AND  AI_ANALOGY('BOXWEL3'    USING MODEL COLUMN PRIM_AUTHOR,
                  'IQATW001'   USING MODEL COLUMN REF_TABLE  ,
                  'KKKKKKK'    USING MODEL COLUMN PRIM_AUTHOR,
                  REF_TABLE )                                 
       > 0.5                                                  
ORDER BY 1 DESC -- SHOW BEST FIRST                            
--ORDER BY 1 -- SHOW WORST FIRST                              
FETCH FIRST 2000 ROWS ONLY ;                                  

And the results:

---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+-------
               AI_VALUE  WLX_TIMESTAMP                               STMT_ID  STMT_TIMESTAMP              PRIM_AUTHOR
---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+-------
+0.7875136583708362E+00  2022-12-16-11.43.46.365129                    54801  2023-01-05-14.23.34.706136  KKKKKKK    
+0.7875136583708362E+00  2022-12-16-11.43.46.365129                    54800  2023-01-05-14.23.34.704234  KKKKKKK    
+0.7875136583708362E+00  2022-12-16-11.43.46.365129                    43654  2023-01-05-08.09.18.888198  KKKKKKK    
+0.7875136583708362E+00  2022-12-16-11.43.46.365129                    43653  2023-01-05-08.09.18.832308  KKKKKKK    
+0.7754887840772942E+00  2022-12-05-11.43.10.984618                     2616  2022-12-05-10.15.20.884139  KKKKKKK    
+0.7754887840772942E+00  2022-12-05-11.43.10.984618                     2609  2022-12-05-10.15.20.487031  KKKKKKK    
+0.7754887840772942E+00  2022-12-05-11.43.10.984618                     2617  2022-12-05-10.15.20.905752  KKKKKKK    
+0.7754887840772942E+00  2022-12-16-11.43.46.365129                    37239  2023-01-04-11.07.49.907438  KKKKKKK    
+0.7754887840772942E+00  2022-12-16-11.43.46.365129                    37230  2023-01-04-11.07.49.457189  KKKKKKK    
+0.7754887840772942E+00  2022-12-16-11.43.46.365129                    37237  2023-01-04-11.07.49.879457  KKKKKKK    
+0.7754887840772942E+00  2022-12-16-11.43.46.365129                    37238  2023-01-04-11.07.49.903769  KKKKKKK    
+0.7754887840772942E+00  2022-12-05-11.43.10.984618                     2618  2022-12-05-10.15.20.909552  KKKKKKK    
+0.7754887840772942E+00  2022-12-16-11.43.46.365129                    45396  2023-01-05-08.17.11.633089  KKKKKKK    
+0.7754887840772942E+00  2022-12-16-11.43.46.365129                    45389  2023-01-05-08.17.11.311055  KKKKKKK    
+0.7754887840772942E+00  2022-12-16-11.43.46.365129                    45397  2023-01-05-08.17.11.655514  KKKKKKK    

+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+--
PROGRAM   REF_TABLE                        EXECUTIONS          GETP_OPERATIONS              ELAPSE_TIME
+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+--
SEDBTIAA  R510T002                                  1                        0                        8
SEDBTIAA  R510T002                                  1                        0                        9
SEDBTIAA  R510T002                                  1                        0                       11
SEDBTIAA  R510T002                                  1                        3                       61
SEDBTIAA  IQATA001                                  0                        0                        0
SEDBTIAA  IQATA001                                  0                        0                        0
SEDBTIAA  IQATA001                                  0                        0                        0
SEDBTIAA  IQATA001                                  1                        4                       32
SEDBTIAA  IQATA001                                  1                        2                      111
SEDBTIAA  IQATA001                                  1                       12                    20749
SEDBTIAA  IQATA001                                  1                        4                       36
SEDBTIAA  IQATA001                                  0                        0                        0
SEDBTIAA  IQATA001                                  1                       12                    18571

All interesting stuff! I use dynamic SQL to INSERT into the table a lot, and it has determined that use of dynamic SQL with tables R510T002 and IQATA001 is analogous. In fact, it is! The SQLs were all INSERT, DELETE and UPDATE… Clever ol’ AI!

Dynamic Duo?

Now I wish to see which programs process dynamic SQL like the IBM DSNTIAD and DSNTIAP programs:

SELECT AI_SEMANTIC_CLUSTER( PROGRAM,             
                           'DSNTIAD',            
                           'DSNTIAP') AS AI_VALUE
      ,A.WLX_TIMESTAMP                           
      ,A.STMT_ID                                 
      ,A.STMT_TIMESTAMP                          
      ,SUBSTR(A.PROGRAM , 1 , 8) AS PROGRAM      
      ,A.EXECUTIONS                              
      ,A.GETP_OPERATIONS                         
      ,A.ELAPSE_TIME                             
      ,A.CPU_TIME                                
      ,A.STMT_TEXT                               
FROM IQA061QB.IQATW001 A                         
WHERE A.PROGRAM NOT IN ('DSNTIAD', 'DSNTIAP')    
  AND A.STMT_ORIGIN = 'D'                        
ORDER BY 1 DESC -- SHOW BEST FIRST                      
--ORDER BY 1 -- SHOW WORST FIRST                            
FETCH FIRST 10 ROWS ONLY ;                       

And the results:

---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+----
               AI_VALUE  WLX_TIMESTAMP                               STMT_ID  STMT_TIMESTAMP              PROGRAM 
---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+----
+0.7104441523551941E+00  2023-01-06-05.27.28.779825                      824  2023-01-17-09.13.48.940697  DSN§EP2L
+0.5050856471061707E+00  2023-01-06-05.27.28.779825                      559  2023-01-12-13.04.44.032345  O2DB81  
+0.5032740235328674E+00  2023-01-06-05.27.28.779825                      561  2023-01-12-13.04.44.075465  O2DB84  
+0.5007491707801819E+00  2023-01-06-05.27.28.779825                      560  2023-01-12-13.04.44.062180  O2DB82  
+0.4917877912521362E+00  2023-01-06-05.27.28.779825                      558  2023-01-12-13.04.43.986695  O2DB80  
+0.4652681946754456E+00  2023-01-06-05.27.28.779825                      562  2023-01-12-13.04.44.105563  O2DB85  
+0.4551711678504944E+00  2023-02-02-10.45.26.535375                        8  2023-01-17-16.51.06.825629  O2DB8X  
+0.4551711678504944E+00  2023-01-06-05.27.28.779825                        8  2023-01-17-16.51.06.825629  O2DB8X  
+0.4551711678504944E+00  2023-01-06-05.27.28.779825                      557  2023-01-12-13.04.40.846826  O2DB8X  
+0.4452087283134460E+00  2023-01-06-05.27.28.779825                        7  2023-01-17-16.50.54.118774  O2DB6X  

+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+-
EXECUTIONS          GETP_OPERATIONS              ELAPSE_TIME                 CPU_TIME  STMT_TEXT                                    
+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+-
         1                        0                        0                        0  INSERT INTO KKKKKKK.SAXDBT (SELECT A.NAME, 'B
         1                        0                       29                       29  SELECT COUNT(*),COALESCE(SUM(CASE EXCEPTION_C
         1                        0                       28                       24  SELECT COUNT(*),COALESCE(SUM(CASE EXCEPTION_C
         1                        0                       22                       22  SELECT COUNT(*),COALESCE(SUM(CASE EXCEPTION_C
         1                        0                       76                       43  SELECT COUNT(*),COALESCE(SUM(CASE EXCEPTION_C
         1                        0                       44                       24  SELECT COUNT(*),COALESCE(SUM(CASE EXCEPTION_C
         0                        0                        0                        0  SELECT COALESCE(COALESCE(A.DBNAME,B.DBNAME),C
         1                      215                   132887                    13462  SELECT COALESCE(COALESCE(A.DBNAME,B.DBNAME),C
         1                       18                     1035                      791  SELECT COALESCE(COALESCE(A.DBNAME,B.DBNAME),C
         2                        7                    27753                     1236  SELECT COALESCE(COALESCE(A.DBNAME,B.DBNAME),C

Again, very nice – it spotted all of the RealTime DBAExpert Dynamic SQL access programs in use…

Undynamic Duo?

Ok, now the opposite of that query, show me the SQLs that are like them but not them!

SELECT AI_SEMANTIC_CLUSTER( PROGRAM,                      
                           'DSNTIAD',                     
                           'IQADBACP',                    
                           'SEDBTIAA') AS AI_VALUE        
      ,A.WLX_TIMESTAMP                                    
      ,A.STMT_ID                                          
      ,A.STMT_TIMESTAMP                                   
      ,SUBSTR(A.PRIM_AUTHOR , 1 , 8) AS PRIM_AUTHOR       
      ,SUBSTR(A.PROGRAM , 1 , 8) AS PROGRAM               
      ,SUBSTR(A.REF_TABLE , 1 , 18) AS REF_TABLE          
      ,A.EXECUTIONS                                       
      ,A.GETP_OPERATIONS                                  
      ,A.ELAPSE_TIME                                      
      ,A.CPU_TIME                                         
      ,A.STMT_TEXT                                        
FROM IQA061QB.IQATW001 A                                  
WHERE A.PROGRAM NOT IN ('DSNTIAD', 'IQADBACP' ,'SEDBTIAA')
--AND A.STMT_ORIGIN = 'D'                                 
--ORDER BY 1 DESC -- SHOW BEST FIRST                      
ORDER BY 1 -- SHOW WORST FIRST                            
FETCH FIRST 10 ROWS ONLY ;                                

and the output:

+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+-------
STMT_ID  STMT_TIMESTAMP              PRIM_AUTHOR  PROGRAM   REF_TABLE                        EXECUTIONS          GETP_OPERATIONS
+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+-------
  39974  2023-01-11-09.15.29.336208               DSMSUMA   SYSVOLUMES                                3                        9
  39973  2023-01-11-09.15.29.336151               DSMSUMA   SYSDATABASE                               3                        6
  39976  2023-01-11-09.15.29.363187               DSMSUMA   SYSTABLEPART                              3                      408
  39975  2023-01-11-09.15.29.343490               DSMSUMA   SYSTABLES                                 3                      870
  39981  2023-01-11-09.15.29.459879               DSMSUMA   SYSPLAN                                   3                        9
  39980  2023-01-11-09.15.29.449311               DSMSUMA   SYSTABLESPACE                             3                      399
  39979  2023-01-11-09.15.29.381912               DSMSUMA   SYSINDEXES                                3                     1970
  39978  2023-01-11-09.15.29.370762               DSMSUMA   SYSINDEXES                                3                      786
  39977  2023-01-11-09.15.29.369892               DSMSUMA   SYSINDEXPART                              3                      381
  39972  2023-01-11-09.15.29.336020               DSMSUMA   SYSSTOGROUP                               3                        6

Aha! It found a little assembler program that fires off SQL like the top three!

The Apple doesn’t Fall far from the Tree

Finally, I want to see which programs behave like IQADBACP (our main dynamic SQL driver program):

SELECT AI_SIMILARITY( PROGRAM,              
                     'IQADBACP') AS AI_VALUE
      ,A.WLX_TIMESTAMP                      
      ,A.STMT_ID                            
      ,A.STMT_TIMESTAMP                     
      ,SUBSTR(A.PROGRAM , 1 , 8) AS PROGRAM 
      ,A.EXECUTIONS                         
      ,A.GETP_OPERATIONS                    
      ,A.ELAPSE_TIME                        
      ,A.CPU_TIME                           
      ,A.STMT_TEXT                          
FROM IQA061QB.IQATW001 A                    
WHERE NOT A.PROGRAM = 'IQADBACP'            
  AND A.STMT_ORIGIN = 'D'                   
ORDER BY 1 DESC -- SHOW BEST FIRST                      
--ORDER BY 1 -- SHOW WORST FIRST                            
FETCH FIRST 10 ROWS ONLY;                   

And the output:

---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---
               AI_VALUE  WLX_TIMESTAMP                               STMT_ID  STMT_TIMESTAMP              PROGRAM
---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---
+0.4575602412223816E+00  2023-02-02-10.45.26.535375                        7  2023-01-17-16.50.54.118774  O2DB6X 
+0.4575602412223816E+00  2023-01-06-05.27.28.779825                        7  2023-01-17-16.50.54.118774  O2DB6X 
+0.4400676488876343E+00  2023-01-06-05.27.28.779825                      220  2023-01-20-10.11.14.618038  DSMDSLC
+0.4400676488876343E+00  2023-01-06-05.27.28.779825                      222  2023-01-20-10.11.38.136712  DSMDSLC
+0.4400676488876343E+00  2023-01-06-05.27.28.779825                      221  2023-01-20-10.11.21.993833  DSMDSLC
+0.4400676488876343E+00  2023-01-06-05.27.28.779825                      252  2023-01-20-10.55.07.078652  DSMDSLC
+0.4400676488876343E+00  2023-01-06-05.27.28.779825                      251  2023-01-20-10.54.37.901247  DSMDSLC
+0.4400676488876343E+00  2023-01-06-05.27.28.779825                      233  2023-01-20-10.47.23.961076  DSMDSLC
+0.4400676488876343E+00  2023-01-06-05.27.28.779825                      232  2023-01-20-10.46.59.756430  DSMDSLC
+0.4400676488876343E+00  2023-01-06-05.27.28.779825                      224  2023-01-20-10.33.42.609175  DSMDSLC

+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+-
EXECUTIONS          GETP_OPERATIONS              ELAPSE_TIME                 CPU_TIME  STMT_TEXT                                    
+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+---------+-
         0                        0                        0                        0  SELECT COALESCE(COALESCE(A.DBNAME,B.DBNAME),C
         2                        7                    27753                     1236  SELECT COALESCE(COALESCE(A.DBNAME,B.DBNAME),C
         2                     1387                    57974                    14900  SELECT CASE WHEN B.VCATNAME < ' ' THEN '00000
         6                     4170                    68943                    53330  SELECT CASE WHEN B.VCATNAME < ' ' THEN '00000
         6                     4596                   286233                    99773  SELECT CASE WHEN B.VCATNAME < ' ' THEN '00000
         1                      851                    55367                    42542  SELECT CASE WHEN B.VCATNAME < ' ' THEN '00000
         1                      298                   122961                    24848  SELECT CASE WHEN B.VCATNAME < ' ' THEN '00000
         2                     1260                    68272                    48952  SELECT CASE WHEN B.VCATNAME < ' ' THEN '00000
         1                      192                     3395                     2508  SELECT CASE WHEN B.VCATNAME < ' ' THEN '00000
         3                      810                    43520                    23771  SELECT CASE WHEN B.VCATNAME < ' ' THEN '00000

Again, it found all of the correct programs.

Quibble Time!

I did find some small problems…

I use ALIASes a lot and they appear in the drop-down selection box when in “Add object”, but if you choose one as an AI Object:

This then leads on to the second quibble… The red windowed error messages stay there until you click them away… This can lead you to believe that a problem exists when in reality everything is groovy!

I also found out that the spinning wheel completes if you wait for Spark and LOAD and then go back and forward on the panel.

Finally, the way you move around the product is a bit odd… sometimes you use the browser back function, sometimes you click on a “Back” button, sometimes you click on a bread crumb, sometimes there are multiple options hidden under triple vertical dots which change depending on where you are in the process.

I am sure these little UI bugs will all get ironed out very quickly!

End of Quibbles.

First Baby Steps Taken!

This little trip into the AI world is really just the tip of the iceberg. I will be doing many more AI queries over the coming months, and I hope to show all my results, either here or in another one of my Newsletters and/or at the German GUIDE in April 2023 and, hopefully, at the IDUG 2023 as well.

Any questions about AI, do not fear to ask, and when not me then ChatGPT!

TTFN

Roy Boxwell

Ok, ok, I am a little bit of a geek… But in my defense at least I have started kicking around with Artificial Intelligence and not just ChatGPT! This month, I wish to dip my toes into the icy, cold waters of AI and show you what you can do in Db2 13 right out-of-the-box !

What’s in a BiF?

Db2 13 FL500 brings three new Scalar BIFs. These are the SQL Data Insights functions. They come supplied with Db2 13 but you do have to install a bunch of stuff to actually get them working (so not really out-of-the-box, but close!)

Five Easy Steps?

First, you need to make sure you have all the prereqs in place. These are basically a couple of APARs for the IBM Z Deep Neural Network Library (zDNN), the z/OS Supervisor, IBM OpenBLAS, z/OS OpenSSH and IBM 64-bit SDK for z/OS Java. zDNN and OpenBLAS come with z/OS 2.4/2.5, but without the required APARs the libraries may be empty.

SQL Data Insights (SQL DI) is a kind of no-charge add-on to Db2 13, so you need to order and install it separately (FMID HDBDD18).

All the prereqs are listed in the Db2 docu: Preparing SQL Data Insights installation

Now the Install

Then you need to install and customize SQL DI, starting with the definition of a (technical) user along with its appropriate authorization (Configuring setup user ID for SQL Data Insights). They’re asking for 100 GB of storage for the zFS home directory, but I think you’ll only need that when you start to run AI model training on vast amounts of data. For my first tiny steps into the world of Db2 AI it worked with a tenth of that without any problems. It may well change with my soon upcoming tests! The requirements listed for CPU and system memory aren’t much smaller and I’m experiencing a very measurable CPU consumption whenever the model training on an object starts.

RACF for Advanced Users!

The next step (Configuring user authentication for SQL Data Insights) is very likely a task for your RACF colleague, unless you have RACF SPECIAL authority (Who on earth *ever* has SPECIAL these days … any auditor would throw a right royal wobbly there!) or sufficient authority as described in RACDCERT command. RACDCERT manages RACF digital certificates and SQL DI needs that to allow secure (https) connections to the user interface, coming as an easy-to-use web application.

While You are Waiting, Sir…

While your RACF colleague is getting the (technical) user and the certificate in place, you can sneak into Db2’s SDSNSAMP lib to customize and execute DSNTIJAI. This guy creates the required database and pseudo-catalog tables as described in Configuring Db2 for SQL Data Insights. There are also GRANTs in the sample job, but I had to add another one for procedure DSNWLM_UTILS, since SQL DI uses that for LOAD.

And We are Finished!

Finally, you must do the SMP/E installation of SQL DI, followed by executing the installation script in the USS environment as described in Installing and configuring SQL Data Insights. USS scripts seem to be (along with certificates) the fun part of installing products these days. Carefully plan, and stick with, the values that you enter during the interactive installation of the sqldi.sh script. If you re-run the installation, for example, and decide for another SQLDI_HOME, your .profile will have a # Generated by SQL Data Insights installation script section that will not be updated. The script also starts SQL DI and SPARK (needed and installed by SQL DI). However, there seem to be some very low internal timeout values set for verifying the successful start – at least in my environment. The script complained that start failed, but it was all up and running fine. After you verified the successful installation as described at Verifying the installation and configuration of SQL Data Insights, you can start having fun with SQL DI. I, however, decided to additionally go for the optional step Creating a started task for the SQL Data Insights application. If you intend to use SQL DI more than once, and maybe have your colleagues also work with it, I think this piece is a must. Be aware that there are a couple of adjustments to make it work:

• The sample STDENV, as well as the samples that are in the SQLDAPPS STC sample job don’t have all the definitions of the .profile sample, which I added manually.
• The AI model training failed, complaining that zade wasn’t found in java.library.path and I was only able to fix that by manually setting additional environment variables as described in IBM Z Artificial Intelligence Data Embedding Library environment
• The _CEE_RUNOPTS=”…” sample didn’t work for me and I had to remove the quotation marks to make it look like _CEE_RUNOPTS=FILETAG(AUTOCVT,AUTOTAG) POSIX(ON)
• Starting/Stopping SQLDAPPS using SQLDAPPS sample job triggers sqldi.sh with the start/stop option. This is accomplished by two members in the PDS specified by the STDPARM DD card. The STOP is issued by command /s SQLDAPPS,OPTION=’SQLDSTOP’ (for what ever reason not by STOP SQLDAPPS) and correctly refers to member SQLDSTOP. The START is issued by command /s SQLDAPPS,OPTION=’STRT’. However, the member is actually called SQLDSTRT, so it either requires to change the STCs sample job default option to SQLDSTRT, or the member to be renamed as STRT.

There is also an optional step to create an STC for the SPARK cluster in the docu (Creating started tasks for the Spark cluster). Short story: Skip it, because SQL DI will start the SPARK cluster (master and worker) automatically anyways.

Here is a link containing a really good overview:

https://developer.ibm.com/articles/use-ibm-db2-sql-data-insights-to-uncover-hidden-relationships-in-your-data/

Up and Running?

Once installed and ok, you can then kick off the web interface. The first page is a login panel of course:

Time to get Connected

Here you can see that I have already defined my little test Db2 13 to the system:

If you click on “Add connection” you get a pop-up to define how to get to another Db2 using a certificate or a userid and password, very similar to setting up a Data Studio connection. Click on the three dots and you get a drop-down box:

Here you can disconnect, Edit (Which shows you the same window as “add connection”), List AI objects or Delete.

Starting to Feel Intelligent!

Choosing List AI objects you see what I have created:

I’m Sorry, Dave, I’m Afraid I can’t do That.

Sadly, it failed – Click on the downward arrow expands the view:

When I first heard “zade” I thought of Sade – “Smooth Operator” for some reason…The explanation is actually a hint that something is wrong with the JAVA definitions. It could be a classpath problem or a version problem. This is in work as I write this!

Can you hear me, HAL?

Now the Enable AI query I will cover later, first Analyze data takes you to the next window:

Just the Facts, Ma’am

Click on Data statistics for a different view:

Column influence:

Has no function yet as the training failed…

An Overview is Good

Selecting View model shows you the history of this model:

Naturally, Cluster center shows this:

Is There an Export Limit?

Export Columns creates a JSON file:

Which then looks like:

Pretty horrible if you ask me!

Skipping back to HAL

Going back to the List AI Objects window, there are two blue buttons: Add object and Run query.

Add Object is how I added my first table, it is basically a catalog browser where you can pick the table of choice and, once selected, the greyed-out boxes at the bottom turn blue and you may click on Add object or Enable AI query. Having added a table you then do indeed click on Enable AI query to decide which columns have which DI data type. This is the most critical moment in the whole process!

When you select a column, you can decide on a data type:

Categorical, Numeric or Key. Once you have decided this for all of the columns click on Next:

Here you get the Filter chance:

Last Chance to Turn Back…

Now you have the choice to filter again or just click on Enable:

This now kicks off an Apache Spark process in the background that sucks all of the CPU and memory from your system for a while and builds a new vector table in the background. The vector table actually contains the model data to be used by the new BiFs.

In SPUFI you can find your vector table:

It is “AIDB_” concatenated with your table creator an underscore and then table name.

There are only three columns and, naturally, due to the failure of the model training, my vector table is empty… The COLUMN_NAME here is the column name that is referred to from now on in this blog.

Round up the Usual Suspects!

The vector data tables are very, very special… you cannot simply copy them between systems like normal data and you must make sure they stay in step with their partner “real data” tables – and here I specifically mean image copies!

Once a model is trained, you can use the new BiFs. The documentation is a bit sparse and it does not explicitly say you can use numeric data types, but it also does not explicitly say you cannot! In fact, the list of “cannot use types” are always the same:

BINARY, VARBINARY, CLOB, BLOB, DBCLOB, XML or ROWID.

Maximum length of the used column is 1868 Bytes.

Where’s the Beef?

Here are the currently available AI BiFs with examples taken from the docu.

This is the human language equivalent of Source-1 is to Target-1 as Source-2 is to Target-2.

The Column Name is the identifier which points to the model and column name to be used for this AI function (COLUMN_NAME in the vector table) and, if not given, then the expression determines the column name or it is just the actual table column name. The model specified must, obviously, be the same for both Sources and the same for both Targets.

Here’s an example showing the syntax:

The customer with ID ‘1066_JKSGK’ has churned. Given the relationship of that customer to ‘YES’ in the churn column, find customers with the same relationship to ‘NO’ in the churn column, in other words, customers unlikely to churn.

The result is a double precision floating point number. The larger the positive value the better the analogy and a value of -1 is a poor analogy. Caution must be used as the result can also be NULL if any of the args are NULL.

This returns a clustering score for the member-expr value among the cluster of values defined in the, up to three times repeated, clustering expression list. Like AI_ANALOGY the model to be used is either the member-expr column name or the explicit column name.

Here’s an example showing the syntax:

Customers with IDs ‘0280_XJGEX’, ‘6467_CHFZW’ and ‘0093_XWZFY’ have all churned. If we form a semantic cluster of those three customers, find the top 5 customers that would belong in that cluster.

The result is a double precision floating point number between -1.0 and +1.0 that is the semantic clustering score. A larger positive number indicates a better clustering among the list of clusters. Caution must be used, as the result can also be NULL if any of the arguments are NULL or were not seen during training.

This returns a similarity score for the two expressions. Like AI_ANALOGY, the model to be used is either the expression column or the explicit column name.

Here are two examples showing the syntax:

Find the top five customers by ID most similar to the customer with ID ‘3668-QPYBK’.

Find the top three payment methods most similar to ‘YES’ in the CHURN column.

The result is a double-precision floating point number (FLOAT) that is the similarity score between -1.0 and 1.0, where -1.0 means that the values are least similar, and 1.0 means that they are most similar.

With this BiF you can get NULL returned if the columns of interest are numeric types but the value is outside of the range of FLOAT. Further, NULL is returned if any argument is NULL and also if the value was not available during training and the column is not numeric.

Follow the Money!

This is all you get out-of-the-box in Db2 13 – The real question is, “For which business use cases does it make sense to use these BiFs?” That is naturally a very hard question and next month I hope to bring you real life AI examples from my test data (as long as I can get my Models trained!)

TTFN

Roy Boxwell

Hi all! Welcome to the end-of-year goody that we traditionally hand out. This year is a relaunch of the Migration HealthCheck that we first did over two years ago. I’ve also provided some news about Db2 13 UTS PBR RPN spaces that might be of interest to you!

DB2 12 UTS PBR RPN

You might well know that this was my single favorite feature of Db2 12. What I did not really appreciate, until now, was the fact that getting these very big partitions can come with a major price!

WHAT IS THE PROBLEM?

If you have a Data Sharing system (Who does not these days?) and you happen to use LOCKSIZE ROW on your UTS PBR RPN then you should take a good look at your performance monitor data. If you see a high number of P-Locks, and the number of false contentions is greater than your IRLM–SUSPENDS, then BINGO!

THE FIX IS?

Documented in the red book „Db2 13 Performance Topics“, Chapter 5 Data Sharing, 5.1 „Partition-by-range table space relative page numbering enhancements“ is the information above, and a lot more. The crux of the matter is a new Hash Algorithm, and to get to it you must simply REORG any UTS PBR RPN spaces that were created prior to Db2 13 FL500.

HERE IS SOME SQL

Here is some SQL to list out any and all of your UTS PBR RPN table partitions that were created prior to Db2 13 FL500 and have not yet been REORGed or LOAD REPLACED.

First check that you are actually *at* Db2 13 R1 FL500!

SELECT LU.EFFECTIVE_TIME 
FROM SYSIBM.SYSLEVELUPDATES LU 
WHERE 1 = 1 
  AND LU.FUNCTION_LVL   = 'V13R1M500'
  AND LU.OPERATION_TYPE = 'F' 
WITH UR 
FOR FETCH ONLY 
; 

This simply returns the time when the FL500 was „activated“ in your Db2 13 system. If it returns no rows then you cannot do anything…

THE USUAL SUSPECTS…

Then we get the Partitions of interest:

SELECT SUBSTR(TP.DBNAME , 1 , 8) AS DBNAME 
      ,SUBSTR(TP.TSNAME , 1 , 8) AS TSNAME 
      ,TP.PARTITION 
      ,TP.CREATEDTS 
      ,TP.REORG_LR_TS 
FROM SYSIBM.SYSTABLESPACE TS 
    ,SYSIBM.SYSTABLEPART  TP 
WHERE 1 = 1 
  AND TS.DBNAME      = TP.DBNAME 
  AND TS.NAME        = TP.TSNAME 
  AND TS.TYPE        = 'R' -- UTS PBR ONLY 
  AND TS.LOCKRULE    = 'R' -- ROW LEVEL LOCKING ONLY 
  AND TP.PAGENUM     = 'R' -- UTS PBR RPN ONLY 
  AND TP.CREATEDTS   < (SELECT LU.EFFECTIVE_TIME 
                        FROM SYSIBM.SYSLEVELUPDATES LU 
                        WHERE 1 = 1 
                          AND LU.FUNCTION_LVL = 'V13R1M500' 
                          AND LU.OPERATION_TYPE = 'F' 
                       )   -- CREATED BEFORE FL500 ACTIVATED 
  AND TP.REORG_LR_TS < (SELECT LU.EFFECTIVE_TIME 
                        FROM SYSIBM.SYSLEVELUPDATES LU 
                        WHERE 1 = 1 
                          AND LU.FUNCTION_LVL = 'V13R1M500' 
                          AND LU.OPERATION_TYPE = 'F' 
                       )   -- LAST REORG/LOAD BEFORE FL500 ACTIVATED
ORDER BY 1 , 2 , 3 
WITH UR 
FOR FETCH ONLY 
; 

This query uses the effective timestamp, created timestamp and the last reorg load replace timestamp to filter out all the partitions that do not need to be REORGed or LOAD REPLACEd. 

Please also remember you only need to do all this when you have ROW LEVEL locking in data sharing with high CPU p-locks. The red book shows some very impressive CPU savings!

BACK TO THE HOLIDAYS!

Over the last two years we have added and enhanced our Migration HealthCheck a lot. Improvements include testing what happens with DEFINE NO spaces when  they were created years and releases ago but would now be externalized. All of this means the output has changed, of course.

TELL ME MORE!

Here’s a list of all the deprecated (and semi-deprecated) items that should be checked and verified at your site:

1. Use of SYNONYMS
2. Use of HASH objects
3. Use of segmented spaces
4. Use of classic partitioned objects (not using table based partitioning)
5. Use of simple spaces
6. Use of six-byte RBA
7. Use of BRF
8. Use of LARGE objects (This is semi-deprecated)
9. SQL EXTERNAL Procedures
10. UNICODE (VARBIN Columns)
11. Old RLF table defs
12. Old PLAN_TABLE defs
13. Old bound packages in use in the last 548 days
14. Direct bound DBRMs (Yes they can still exist!) 

ANYTHING ELSE?

Well yes! You could also check how many empty implicit databases and how many empty tablespaces you have. While you are scanning your subsystem, it could also be cool to list out all the Db2 subsystem KPIs. What about seeing how many tables you actually have in multi-table tablespaces that, at some point, must also be migrated off into a UTS PBG or UTS PBR tablespace?

WE DO IT ALL!

Our little program does all of this for you. It runs through your Db2 Catalog in the blink of an eye and reports all of the data mentioned above.

WHAT DOES IT COST?

Nothing – It is our freeware for 2022/2023 and you only have to be registered on our website to request it along with a password to run it.

HOW DOES IT LOOK?

Here is an example output from one of my test systems here in Düsseldorf:

Db2 Migration HealthCheck V2.3 for SC1 V12R1M510 started at  
                   2022-12-14-10.56.00
           Lines with *** are deprecated features 

Number of DATABASES            :       594 
  # of empty DATABASES         :       237 
  # of implicit DATABASES      :       385 
  # of empty implicit DATABASES:       207 

Number of TABLESPACES          :      4861 
  of which HASH organized      :         0 
  of which PARTITIONED CLASSIC :         2 ***
    # Partitions               :        32 ***
  of which SEGMENTED           :       294 ***
  of which SIMPLE              :         0 
  of which LOB                 :        67 
  of which UTS PBG             :      4467 
    # Partitions               :      4477 
  of which UTS PBR (Absolute)  :         5 
    # Partitions               :       801 
  of which UTS PBR (Relative)  :         6 
    # Partitions               :       756 
  of which XML                 :        20 

Number of tablespaces as LARGE :         8 *** 
Number of empty tablespaces    :        28 
Number of multi-table TSs      :        55 
  # of tables within these     :       239 
Number of incomplete TS        :         1 XXX
Number of INSERT ALG 0 TS      :      4861 
Number of INSERT ALG 1 TS      :         0 
Number of INSERT ALG 2 TS      :         0 

Number of tables               :     10293 
  of which ACCELERATOR ONLY    :         0 
  of which ALIASes             :      5307 
  of which ARCHIVEs            :         1 
  of which AUXs                :        60 
  of which CLONEs              :         0 
  of which GTTs                :       136 
  of which HISTORYs            :         1 
  of which MQTs                :         1 
  of which TABLEs              :      4765 
  of which VIEWs               :         2 
  of which XMLs                :        20 
Number of tables with Audit    :       101 
Number of tables with Data Cap :         0 
Number of tables incomplete    :         1 XXX
Number of tables with control  :         0 

Number of RLF DSNRLMT__ tables :         0 
  of which columns deprecated  :         0 
Number of RLF DSNRLST__ tables :         1 
  of which columns deprecated  :         0 
 
Number of PLAN_TABLES          :        68 
  of which deprecated          :         3 ***

Number of SYNONYMs             :         1 *** 

Number of UNICODE V11 Columns  :         0 

Number of PROCEDURES           :       110
  of which SQL EXTERNAL        :         0
  of which EXTERNAL            :       108
  of which NATIVE SQL          :         2

Number of FUNCTIONS            :        87
  of which EXTERNAL TABLE      :        38
  of which EXTERNAL SCALAR     :        42
  of which SOURCED AGGREGATE   :         0
  of which SOURCED SCALAR      :         0
  of which SQL TABLE           :         1
  of which SQL SCALAR          :         6
  of which SYSTEM-GENERATED    :         0

Number of Indexes              :     23243 
  of which HASH                :         0 
  of which type 2              :     23210 
    # of partitioned IXs       :         6 
    # Partitions               :       160 
  of which DPSI                :        18 
    # Partitions               :       164 
  of which PI                  :        15 
    # Partitions               :      1138 
Number of indexes COPY YES     :        38
Number of indexes COMPRESS YES :         0

Number of table partitions     :      6606 
  of which DEFINE NO           :      2848 
  of which six byte RBA <11 NFM:         0 
  of which six byte RBA Basic  :         0 
  of which ten byte RBA        :      3759 
Number of TP in BRF            :         0
Number of TP with COMPRESS Y   :       498
Number of TP with COMPRESS F   :         0
Number of TP with COMPRESS H   :         0
Number of TP with TRACKMOD YES :      2968

Number of index partitions     :     24666 
  of which DEFINE NO           :     20140 
  of which six byte RBA <11 NFM:         0 
  of which six byte RBA Basic  :         0 
  of which ten byte RBA        :      4527 

Number of STOGROUPS            :        10 
Number of non-SMS VOLUMES      :         0 

Number of PLANs                :        54 
  of which DBRMs direct        :         0
    # of SQL statements        :         0
Number of PACKAGES (total)     :      5788
  of which VALID = A           :        42
  of which VALID = H           :         0
  of which VALID = N           :        44
  of which VALID = Y           :      5702
  of which VALID = S           :         0
  of which OPERATIVE = N       :         0
  of which OPERATIVE = Y       :      5788 

Old RELBOUND executed packages :         0

Number of PACKAGES (distinct)  :       480

Number of Original PACKAGES    :         0
Number of Previous PACKAGES    :         0
Number of Phased-out PACKAGES  :         0
Total number of PACKCOPY       :         0
  of which VALID = A           :         0
  of which VALID = H           :         0
  of which VALID = N           :         0
  of which VALID = Y           :         0
  of which VALID = S           :         0
  of which OPERATIVE = N       :         0
  of which OPERATIVE = Y       :         0
Number of SQL statements       :    441833 

Db2 Migration HealthCheck V2.3 for SC10 V12R1M510 ended at
                   2022-12-14-10.56.03

Db2 Migration HealthCheck ended with RC: 0 

Any line with *** at the end means that you have something to do at some point in the future.  The names of all the found objects are written to DD card DEPRECAT so you can then start building a „to do“ list. I would start now to slowly „fix“ all of these before it is 03:00 in the morning, someone is migrating to Db2 14 FL 608 and it all goes horribly wrong…

Any line with XXX means that you have an incomplete definition for a tablespace and/or a table. These should be fixed as well, either by completing the definition or dropping the unfinished object(s).

WHAT’S WRONG WITH LARGE?

This is not actually deprecated but any tablespaces marked as LARGE tend to also not have a valid DSSIZE in them. This is fine if you have built a CASE construct to derive the value from the tablespace definition. But what you should do, is an ALTER and a REORG to „move“ the LARGE to a „proper“ tablespace. IBM and 3rd Party Software vendors hate having to remember that ancient tablespaces are still out there!

ALL ON MY OWN?

Naturally not! For example, after all the ALTERs have been done, a lot of the spaces are simply in Advisory REORG pending status and you could use our RealtimeDBAExpert (RTDX) software to automatically generate the required REORGs to action the changes.

SYNONYMS??

Well, you can do them all yourself by reading one of my older newsletters – just remember to watch out for the GRANTs afterwards.

HOW MANY DEPRECATED OBJECTS DO YOU HAVE?

I would love to get screenshots of the output at your sites which I would then all sum up and publish as an addendum to this newsletter. Just so that people can see how many Parrots we all have pining for the fjords!

TTFN, Happy Holidays!

Roy Boxwell

This month, I am turning over my blog to my colleague Andre Kuerten from Software Engineering’s German Labs based in Düsseldorf, Germany. I challenged him to write a blog all about his experiences as a “first time IDUG EMEA attendee” and all that that entailed as I thought this would make an interesting read for us all!

I’ll be Back!

Have no fear, dear readers, as I will be back next month with our annual Christmas give-away. The first blog of 2023 will be my comprehensive review of the 2022 IDUG EMEA in Edinburgh and the surprises found there-in.

It begins…

IDUG EMEA 2022 – First Timer Report

Pre-Preparation-Phase:

My firm gave me the chance to go the IDUG EMEA, additionally taking a Saturday workshop to educate myself in all things SQL. Surprisingly, my wife gave approval! Therefore, “the guy from techsupp” that I am, planned the trip to Edinburgh with all the hotel and flight bookings etc.

Saturday:

Unexpectedly, everything went smoothly and so I was standing in front of the Edinburgh International Conference Center (EICC) on Saturday morning, ready for my first IDUG in person after working for more than 15 years in the Db2 business, professionally developing software.

The general plan was for me to learn, or refresh my knowledge, about query optimization and finding the cause for poor performing SQL to be ready to work a little bit more intensively in this area. The first decision made was that I will take part in the “Query Optimization and Tuning Workshop” that will take the complete Saturday, covering themes like statistics, cardinalities, optimizer stuff, query EXPLAIN and strategies for performance tuning.

I entered the EICC and was registered by the very friendly and helpful IDUG employees, got my badge and a bag filled with little goodies – I think the practical value of the umbrella was unbeatable.

When looking around I have to admit that the EICC is a nice venue. Right behind the entrance is a big hall on the ground level which opens up for you, they arranged some high tables to put down your water/coffee, just a few more chairs would have been a good idea. But I was impressed by the sheer size of it all.

Sched is Your Friend!

The rooms all had Gaelic names, you just had to locate them on the map, so everything could be easily found. The Sched App was a perfect addition to the printed schedules, it made it very easy to get the day organized and to know where to go next. I really liked it and the connection to the website.

Workshop – z/OS???

Took the escalators to “Carrick” and got a seat in the workshop, looked around and discovered the usual mix of technical geeks, where the average age seems to be a little low… Additionally, there was no work station or material lying around that we would have to work with during the day (I had been told by the experienced colleagues that I would really have to work at the workshop), so I started to wonder… Chatting with some of the people in the room was fine, and I was glad I was not the only first timer (I must admit that I didn’t apply for a first timer badge, I do have my limits). Then it started. Instead of the expected z/OS hands-on workshop it slowly turned into an LUW daily presentation… At least I was not alone, as another z/OS guy was also not expecting LUW. It had not been made clear at early registration time that this was going to be just an LUW workshop. Anyway, since the topic was “SQL”, there was still valuable information here. So I listened and concentrated and got new ideas about how to start with SQL optimization and where to look first, starting from query optimization basics, discussing some database design alternatives and ending with cardinality estimation.

After a small lunch, taken in the big hall, the afternoon session started and was now going for the optimization of various operations like Scans, Sort, INSERT, UPDATE, DELETE and queries with outer joins, aggregation, distinct, correlated subqueries etc. I just had to take care to keep focused while the knowledge was being distributed. I think this would have been much easier if we could have done some practical exercises.

At the end the feelings about the workshop were a little bit mixed, but overall positive.

The Roy arrives

Starting in the late afternoon, I kept getting status messages from Roy (Boxwell, just in case anybody knows him 😊) who was on his way to Edinburgh, so our team size doubled in the evening.

The Booth…

Since my firm was a vendor at the IDUG I also got some experience from building the booth (where I really have to say thanks to the people from the EICC for all their help and kindness) and running it.

Sunday:

On Sunday the first normal IDUG Sessions were starting, I made it to the initial key-note in the “Pentland Main Auditorium” titled “Why Some Teams Are Successful While Others Struggle”, it was interesting, even though it was not a direct technical approach, it was about the people that you are working with and trusting your team. The auditorium (more like a cinema theater) was half full and they mentioned that we were 350+ onsite visitors which I found a good number, however I was told that there used to be many more in the past.

The next session, “Db2 13 for z/OS and More!” was summing up what I already knew from all the announced Db2 13 features, remarkable was one of the presenters, Haakon Roberts, he really did an excellent job and, looking back, I would say that his accent was the best that I heard during IDUG.

Noteworthy is that you soon get used to listening to speakers with different, strange to me, accents very fast, even if you are not a native speaker. This had been one of my fears beforehand which was, luckily, unfounded.

Back to “Carrick” and into the “Back to Basics: High Performance Application Design and Programming” presentation held by Tony Andrews. I mention him because of a reason: As expected the sessions differed in presentation style, quality of presentation etc., this is what you can tick on the evaluation cards. But the best ones, for me, were the ones where you could feel that the speaker was really deeply involved in the work with Db2 and knew what they were talking about like in this session, I heard a lot about the little things like row size, clustering order, all from a practical view, which was all very useful.

“Do I Really Need to Worry about my Commit Frequency? An Introduction to Db2 Logging” confirmed my impression that the commit frequency is something to worry about, but the afternoon highlight was “COBOL abound”, demonstrating how you can develop in COBOL these days, of course, not only on the mainframe , but using Zowe (which I also use and explore at my firm) and the containers you can get there. This was really fascinating, although a little bit special…

Monday:

Day 3, Monday, I also went to the key-note “Behind The Birth Of An Accidental Enterprise”, covering some history of Db2’s evolution.
When talking about the next one, “Getting RID of RID Pool RIDdles”, I have to mention that this was done by two speakers and one of them, Adrian Collett, is known to me, we have worked together with some trial installations.
This was one of the big points: To meet people face to face for the first time, or again after the pandemic, especially customers who I have “talked to” for many years via email or telephone. I got the impression from everyone that they liked it very much to attend in person again.
The presentation itself shed some light on RID pool monitoring and tuning.
Sadly I have to say that this session was one of those that had to “speed up” towards the end (which was not Adrians fault!), so I am really looking forward to getting my hands on the PDFs as some speakers simply ran out of time and then rushed through their presentations. Also the notes taken correspond to slides that I need to see again, so this service (providing the presentations) should be continued for attendees, maybe IDUG could be just a little bit quicker making them available.

The next, very interesting, one was “Access Paths Meet Coding” which gave me practical insights about how to control some basic SQL rules which must be respected and how big the effect of it is in a productive environment.

Tuesday:

Then it was already Tuesday, off to listen to “SMF Records, IFCIDs, Trace Classes – What Does it all Mean?” which was obligatory for me, IFCIDs are utilized broadly in our products.

It should also be mentioned that, in my eyes, the technical equipment provided for the speakers was more than sufficient and the audio control was done by, always available, technical stuff from the congress center making a very good impression.

“Get Cozy with Traces in Db2 for z/OS” reminded me of one way to get worthwhile information about what is going on (or going wrong) on the system.

I have worked with Db2 for some time “in real life” so the next sessions did not have very much new info for me, but I still took notes for “Partitioning Advances: PBR and PBR RPN”, “How to Keep Bad SQL Out of Production” and “Your Statistics are Safe with Me, Statistics Profile Revealed” and have to be checked again.

And the Oscar goes to…

For the evening, the IBM Db2 appreciation event was announced, so we took a walk after the conference day ended up going to the National Museum. The National Museum was, of course, closed when we arrived, so we ended up waiting in the crowd for something to happen. A bagpiper started to play: a perfect introduction for the evening. Worthy of note was the AC/DC part. Finally we made it to the grand hall, perfect location and ambience, food and drinks really well arranged.
Most impressive for me, was the lifetime award for Mr. John Campbell (he had already been honored previously at a keynote). Even I had read a lot from, and about, him (and the respect shown towards him from all of the audience was also good to see). At the end even Roy got an award: he did well as a “newbie” champion. I think they simply forgot to announce him for “some” years. Deserved without question, but why must he get a trophy for his desk that we have to look at (and that is mentioned by him) every single day?

Wednesday:

Back to the IDUG content on the next, and last day, Wednesday:

The last sessions for me at the IDUG were at “Moorfoot”, “Sidlaw” and “Kilsysth” rooms to be visited, following my path of getting new input to solve performance issues, “Db2 SWAT Team Perspective on Db2 13 Features that Maximize Continuous Availability” gave me an outlook to the future at customers sites.

 “The Exciting Journey Towards DevOps on the Mainframe” also explained some “modern” approaches. I regretted, a little bit, that I didn’t go to Roy’s “Esoteric functions in Db2 for z/OS“, instead I took “Explain explained” which was a bit of a basic session, getting the attention back to the little EXPLAIN data details.

At lunch time a, from my point of view, very interesting conference ended, giving me a lot of valuable information and “face-to-face” contacts, some minor negative items are normal I guess, but generally I really appreciated it.

So there you have it!

Many thanks to Andre for writing all that up! Coming soon will be Roy’s take on the EMEA 2022 where I go into technical details about the sessions etc.

I hope you enjoyed the guest blog this month and, as always, let me know what you think!

TTFN

Roy Boxwell & Andre Kuerten

In part two of this newsletter, I wish to bring you up to speed on all the changes in the profile arena from Db2 11 right up until Db2 13.

Filters!

Profiles basically need some sort of “limit” to show Db2 which things should get which profile, and thus which keyword and attribute. To do this, we use Filters; for the filtering data there is an order of preference:

Multiple profiles?

When more than one profile applies to a thread or connection, the evaluation of the different profiles is not simultaneous. Instead, the profiles are evaluated in the following order, according to the criteria that are specified in the profile:

1. IP address or domain name, in the LOCATION column.

2. Product identifier, in the PRDID column.

3. Role and authorization identifier, in both ROLE and AUTHID columns.

4. Role, in the ROLE column only.

5. Authorization identifier, in the AUTHID column only.

6. Server location name, location alias, or database name, in the LOCATION column.

7. The location name of a requester, for monitored threads from a Db2 for z/OS requester. This is only for MONITOR THREADS and MONITOR IDLE THREADS.

8. Collection identifier and package name, in both COLLID and PKGNAME columns.

9. Collection identifier, in the COLLID column only.

10. Package name, in the PKGNAME column only.

11. Client application name, in the CLIENT_APPLNAME column.

12. Client user identifier, in the CLIENT_USERID column.

13. Client workstation name, in the CLIENT_WRKSTNNAME column.

First Come, First Served!

Only the first evaluated applicable profile is applied. Because the evaluation of multiple profiles is not simultaneous, the number of connections, or threads, on the subsystem might change during the evaluation of multiple profiles. Any profile that specifies a specific value in a particular column has precedence over a profile that specifies a single-byte asterisk value (‘*’) in the same column.

Destructive Overlap!

Further, each profile entry cannot have overlapping filter categories. From the precedence list there are eight categories formed from the numbers 1, 2, 3 – 5, 6 – 7, 8 – 10, 11 ,12 and 13, otherwise multiple rows must be inserted.

When you have multiple rows with overlapping filters from different categories then Db2 applies them all. Exact values are higher in priority than wildcard (*). As an example, for product id, PRDID, DSN13011 is before DSN* which is before *.

The More the Merrier!

Db2 also assumes that any rows with more filter values are higher priority than rows with some defaults or NULL values.

Finally, if everything is the same in the filter categories, Db2 will take the last inserted as it assumes this is “the most current version”.

Clear as Mud!

As you can easily see, it is very easy to tie yourself up in knots with this system! Good planning and good testing are paramount to a good, glitch-free implementation!

What Was New in Db2 11?

Well, Db2 11 brought in SPECIAL_REGISTER handling to the profile tables which is especially good for remote accessing threads. So you can now issue SET CURRENT APPLICATION COMPATIBILITY or SET CURRENT PACKAGE PATH, for example.

SET What You Want!

SPECIAL_REGISTER in the KEYWORDS column, ATTRIBUTE1 is any of the accepted SET statements, up to a maximum length of 1024 bytes. ATTRIBUTE2 and ATTRIBUTE3 are both NULL. The filtering is also not case sensitive.

Precedence of the SET special register:

1. Special register explicitly set by the application.

2. Special register set through Profile Support as above.

3. Special register set on the connection property level or data source level.

Buyer Beware!

Db2 11 also introduced warnings about not deleting rows from the _HISTORY tables to make sure you only delete rows that are really gone from the “normal” tables.

Additional Details

The MONITOR CONNECTIONS got a _DIAGLEVEL3 added:

MONITOR CONNECTIONS in the KEYWORDS column, ATTRIBUTE1 is a “two part” column value. The first part is either WARNING or EXCEPTION. A warning causes a console message every five minutes depending on the diagnosis level. An exception issues the diagnosis level and fails the connection request. The second part is either not there or it is_DIAGLEVEL1 which issues a DSNT771I console message, _DIAGLEVEL2 which issues a DSNT772I console message with more details every five minutes at most. _DIAGLEVEL3 issues, for a warning, a DSNT773I console message with more thread details for every thread, and for an exception a DSNT774I console message. ATTRIBUTE2 is a positive integer to indicate the threshold for the maximum number of remote connections. It must be less than or equal to CONDBAT. ATTRIBUTE3 is NULL. Filtering is only by the LOCATION column.

MONITOR IDLE THREADS got three new ATTRIBUTE1 values: EXCEPTION_ROLLBACK aborts any active idle threads and issues DSNT771I, EXCEPTION_ROLLBACK_DIAGLEVEL1 which is the same, and EXCEPTION_ROLLBACK_DIAGLEVEL2 with message DSNT772I.

MONITOR THREADS got _DIAGLEVEL3 added, which for an EXCEPTION issues a DSNT774I console message and, depending on the filtering, can be queued or suspended. WARNING issues the console message DSNT773I for every thread that exceeds the profile threshold.

Updated Info!

The info table about which columns can filter etc. got an overhaul with the note “The value is not case sensitive” for all values for MONITOR CONNECTIONS, MONITOR THREADS, MONITOR IDLE THREADS and SPECIAL_REGISTER.

IP6 Support!

It was also in Db2 11 that IPv6 got supported in the LOCATION field. So, it was now an IP Address: IPv4 dotted-decimal, or an IPv6 colon-hex, or a Domain Name, or a Location name.

What Was New in Db2 12?

In Db2 12, some new KEYWORDS options were introduced:

MONITOR ALL CONNECTIONS in the KEYWORDS column, ATTRIBUTE1 is a “two part” column value. The first part is either WARNING or EXCEPTION. A warning causes a console message every five minutes, depending on the diagnosis level. An exception issues the diagnosis level and fails the connection request. The second part is either not there or it is_DIAGLEVEL1 which issues a DSNT771I console message, _DIAGLEVEL2 which issues a DSNT772I console message with more details every five minutes, at most, and _DIAGLEVEL3 which issues a DSNT773I for WARNING and DSNT774I for EXCEPTION console message with more details. ATTRIBUTE2 is a positive integer to indicate the threshold for the total cumulative number of remote connections from all application servers . It must be less than or equal to CONDBAT. ATTRIBUTE3 is NULL. Filtering is only by the LOCATION column which must contain ‘*’, ‘:::0’ or ‘0.0.0.0’

MONITOR ALL THREADS in the KEYWORDS column, ATTRIBUTE1 is a “two part” column value. The first part is either WARNING or EXCEPTION. A warning causes a console message every five minutes, depending on the diagnosis level. An exception issues the diagnosis level and can cancel the thread, depending on the filtering criteria. Otherwise the thread is queued. The second part is either not there or it is _DIAGLEVEL1 which issues a DSNT771I console message, and _DIAGLEVEL2 which issues a DSNT772I console message with more details. _DIAGLEVEL3 issues a DSNT773I for WARNING and DSNT774I for EXCEPTION console message. ATTRIBUTE2 is a positive integer to indicate the threshold for the total cumulative number of active server threads. It must be less than or equal to MAXDBAT. ATTRIBUTE3 is NULL. Filtering is only by the LOCATION column which must contain ‘*’, ‘:::0’ or ‘0.0.0.0’.

Variable Support

GLOBAL_VARIABLE in the KEYWORDS column, ATTRIBUTE1 is a SET statement for a global variable. E.g. SET SYSIBMADM.GET_ARCHIVE = ‘Y’ or, if you are at Db2 12 FL507 or higher, SET SYSIBMADM.MAX_LOCKS_PER_TABLESPACE = 9000. See the SET documentation in the SQL Reference for more details. ATTRIBUTE2 and ATTRIBUTE3 are both NULL. These are only valid for remote applications.

SHARE_LOCKS in KEYWORDS column, ATTRIBUTE1 column contains a property that applies to global transactions in an RRS context, such as CICS through the External CICS interface. The property applies only to remote applications. The value must be in the following format: PROCEDURE_LIST=aaaa,bbbb,… each of the listed procs cannot be an external SQL proc and not a three part name. Maximum length is 1024 bytes. ATTRIBUTE2 and ATTRIBUTE3 are both NULL.

What Is New in Db2 13

FL500 introduced two extra keywords:

RELEASE_PACKAGE in the KEYWORDS column, ATTRIBUTE1 is COMMIT. ATTRIBUTE2 NULL means this is for remote threads only. 1 applies to local threads only (applies at package load) and 2 profile applies to both local and remote threads. ATTRIBUTE3 is NULL.

SPECIAL_REGISTER in the KEYWORDS column, ATTRIBUTE2 NULL means this is for remote threads only. 1 applies to local threads only (applies at package load) and 2 profile applies to both local and remote threads. Note that only SET CURRENT LOCK TIMEOUT (But not the WAIT, MODE or TO syntax) is currently supported for local threads.

FL 501 introduced local global variable support:

GLOBAL_VARIABLE in the KEYWORDS column, ATTRIBUTE2 NULL means this is for remote threads only. 1 applies to local threads only (applies at package load), and 2 profile applies to both local and remote threads. Note that only SET SYSIBMADM.DEADLOCK_RESOLUTION_PRIORITY = xxx is currently supported for local threads.

The Future Is Bright!

It must be assumed that more and more things will end up in these profiles and it will get more and more interesting to use them, but the major problem is that they are not really transparent. It is very easy to incorrectly set them up and they are tricky, if not impossible, to test. All that being said, they are a very important tool in the tool-box of the modern, agile, DBA!

Examples

Use Case 1 : Evaluate a parameter change

You want to evaluate the impact of a modification at a system or application parameter. You can create a specific profile, with the values you want to activate, and an action level of type WARNING.  After starting the new profile, follow the message DSNT773I to monitor the future impact of your modification, without impacting the subsystem behaviour. 

Use Case 2 : Avoid to adapt attributes of your NULLID Collection

For dynamic SQL, it is not recommended to adapt parameters (APPLCOMPAT, CONCENTRATESTMT …) of the NULLID collection, as it would impact every client working with the default collection.  So, if you want an application to use specific options, you could duplicate the packages of the NULLID collection in a specific collection with the appropriate bind parameters.  Defining a profile that identifies the application, you can redirect to the new COLLID with the use of the special register PACKAGE PATH.  Should your new settings not be optimal, a simple stop of the profile will restore the situation.

Use Case3: High Performance DBAT but *not* everywhere!

The problem:
NULLID being used for all remote access and you wish to use High Performance DBATs for some of them but not all.
The solution:
Create a new collection, called e.g. HIGHPERDBAT, and bind into it any and all of the packages you want available for High Performance DBAT (so also with RELEASE(DEALLOCATE) naturally!)
Insert a Profile Id with a filter for the criteria you wish for, (see earlier in this blog). Then insert an attribute keyword of SPECIAL_REGISTER with ATTRIBUTE1 to be SET CURRENT PACKAGE PATH = ‘HIGHPERDBAT’
If the profile is started, then at next connection time, this new collection will be honored. If something goes awry, and you start getting DBAT problems, simply disable this profile entry and you are done!

Cool huh?

Remember that you must also *allow* High Performance DBATs by running with DDF parameters CMSTAT set to “INACTIVE” and PKGREL set to “BNDOPT” or “BNDPOOL”.

As always, I would love to hear any comments or criticism about this topic!

TTFN

Roy Boxwell

This month I begin a two-part topic because it is just too large to do in one blog entry!

In the Beginning

The DSN_PROFILE_TABLE was introduced sometime in DB2 V8, but it was not until DB2 9 that it started to be used for system profiling when IBM introduced three new commands: DISPLAY PROFILE, START PROFILE and STOP PROFILE. This first appearance of PROFILES was a bit limited and could control only a few ZPARMs – and four of those just for EXPLAIN purposes.

How does/did it look?

To get it working, you must first create all the required tables and indexes. (The DDL is in the db2hlq.SDSNSAMP member DSNTIJSG.) In bold and italics are the DB2 10 and higher versions:

SYSIBM.DSN_PROFILE_TABLE
CREATE TABLE SYSIBM.DSN_PROFILE_TABLE  
       ( "AUTHID"                VARCHAR(128)
        ,"PLANNAME"              VARCHAR(24)
        ,"COLLID"                VARCHAR(128)
        ,"PKGNAME"               VARCHAR(128)
        ,"LOCATION" "IPADDR"     VARCHAR(254)
        ,"PROFILEID"             INTEGER       NOT NULL
            PRIMARY KEY GENERATED BY DEFAULT AS IDENTITY
        ,"PROFILE_TIMESTAMP"     TIMESTAMP     NOT NULL WITH DEFAULT
        ,"PROFILE_ENABLED"       CHAR(1)       NOT NULL DEFAULT 'Y'
        ,"GROUP_MEMBER"          VARCHAR(24)                       
        ,"REMARKS"               VARCHAR(762)                       
        ,"ROLE"                  VARCHAR(128)                      
        ,"PRDID"                 CHAR(8)                           
        ,"CLIENT_APPLNAME"       VARCHAR(255)                      
        ,"CLIENT_USERID"         VARCHAR(255)                      
        ,"CLIENT_WRKSTNNAME"     VARCHAR(255)                      
       );                                                           
CREATE UNIQUE INDEX SYSIBM.DSN_PROFILE_TABLE_IX_ALL
       ON SYSIBM.DSN_PROFILE_TABLE                 
       ( "PROFILEID"                               
       );                                          
CREATE        INDEX SYSIBM.DSN_PROFILE_TABLE_IX2_ALL
       ON SYSIBM.DSN_PROFILE_TABLE                 
       ( "PROFILE_ENABLED"                         
        ,"AUTHID"                                  
        ,"PLANNAME"                                
        ,"COLLID"                                  
        ,"PKGNAME"                                 
        ,"LOCATION" "IPADDR"                        
        ,"PRDID"                                   
        ,"ROLE"                                    
        ,"CLIENT_APPLNAME"                         
        ,"CLIENT_USERID"                            
        ,"CLIENT_WRKSTNNAME"                       
        ,"GROUP_MEMBER"                            
        ,"PROFILE_TIMESTAMP" DESC                  
       );                                          

SYSIBM.DSN_PROFILE_HISTORY – Same columns as DSN_PROFILE_TABLE apart from

REMARKS -> STATUS VARCHAR(254) and no index.

SYSIBM.DSN_PROFILE_ATTRIBUTES
CREATE TABLE SYSIBM.DSN_PROFILE_ATTRIBUTES                         
       ( "PROFILEID"             INTEGER       NOT NULL            
            REFERENCES SYSIBM.DSN_PROFILE_TABLE ON DELETE CASCADE  
        ,"KEYWORDS"              VARCHAR(128)  NOT NULL            
        ,"ATTRIBUTE1"            VARCHAR(1024)                     
        ,"ATTRIBUTE2"            INTEGER                           
        ,"ATTRIBUTE3"            FLOAT                             
        ,"ATTRIBUTE_TIMESTAMP"   TIMESTAMP     NOT NULL WITH DEFAULT
        ,"REMARKS"               VARCHAR(762)                       
       );                                                           
CREATE UNIQUE INDEX SYSIBM.DSN_PROFILE_ATTRIBUTES_IX_ALL           
       ON SYSIBM.DSN_PROFILE_ATTRIBUTES                            
       ( "PROFILEID"                                                
        ,"ATTRIBUTE_TIMESTAMP"   DESC                              
        ,"KEYWORDS"                                                
        ,"ATTRIBUTE1"                                              
        ,"ATTRIBUTE2"                                              
        ,"ATTRIBUTE3"                                              
       );                                                           

SYSIBM.DSN_PROFILE_ATTRIBUTES_HISTORY same columns as DSN_PROFILE_ATTRIBUTES apart from

REMARKS -> STATUS VARCHAR(254)  and no index.

Notice the RI between the DSN_PROFILE_ATTRIBUTES and DSN_PROFILE_TABLE keyed on PROFILEID. Also notice that there are no indexes on the HISTORY tables and also no RI.

So What Could You Do?

With this new functionality you could use a profile to override four ZPARMs, namely NPGTHRSH, OPTIOWGT, STARJOIN and SJTABLES. To do so, you first inserted a row into the DSN_PROFILE_TABLE with some sort of filter, at this time only COLLID and PKGNAME, and then one or more inserts in the DSN_PROFILE_ATTRIBUTES table using the PROFILEID that you either just used, or got generated for you, in the DSN_PROFILE_TABLE using the KEYWORDS column and ATTRIBUTEn column(s).

Always on?

The column PROFILE_ENABLED in the DSN_PROFILE_TABLE informs Db2 whether or not to consider this profile when the START PROFILE command is issued. Setting it to N puts all of this profile’s records “to sleep”.

Not Just ZPARMs

It also enabled three global changes (no filters allowed) for BPname, MAX_RIDBLOCKS and SORT_POOL_SIZE. All of these are just for modelling production systems in test to then get a better, more accurate, EXPLAIN result and have *no* effect on the actual system at all.

Finally, IBM added some Accelerator-only support which had to be done with IBM involved.

Interestingly enough, there was a complete chapter about using profiles to monitor and report on SQL but there was also an update to the docu:

Important: The use of profile tables to monitor and capture information about the performance of SQL statements is deprecated, and not recommended.

So, I will not even bother going into detail about the monitor settings.

What Was the Difference?

The major difference between the SQL and ZPARM settings, was the ability to use different filter column values like AUTHID or IPADDR/LOCATION.

The DSN_PROFILE_HISTORY has the same columns as the DSN_PROFILE_TABLE, except that REMARKS is called STATUS and gets a value set by the START PROFILE command. Basically, a string that starts with REJECTED – or ACCEPTED – and then a text string describing why the profile was, or was not, accepted for use.

What’s in an ATTRIBUTE?

The DSN_PROFILE_ATTRIBUTES table contains the option that should be overridden when the Profile is active and the filtering allows it. The columns of interest are KEYWORDS and the three ATTRIBUTEn columns.

BUFFERPOOL Modelling

BPname (where name is any of the valid names like 0 through 49 or 32K1 through 32K9 etc.) in the KEYWORDS column. ATTRIBUTE1 and ATRIBUTE3 are set to NULL and ATTRIBUTE2 contains a positive integer value for the size of the BUFFERPOOL (for production modelling).

RIDPOOL Modelling

MAX_RIDBLOCKS in the KEYWORDS column, ATTRIBUTE1 and ATTRIBUTE3 are set to NULL, and ATTRIBUTE2 contains a value from 0 to the maximum value that you can set MAXRBLK in that subsystem (for production modelling).

STARJOIN Control

STAR JOIN in the KEYWORDS column, ATTRIBUTE2 and ATTRIBUTE3 are set to NULL, and ATTRIBUTE1 set to DISABLE or ENABLE.

MIN STAR JOIN TABLES in the KEYWORDS column, ATTRIBUTE1 and ATTRIBUTE3 are set to NULL, and ATTRIBUTE2 contains a value from 3 to 225.

INDEX ACCESS Control

NPAGES THRESHOLD in the KEYWORDS column, ATTRIBUTE1 and ATTRIBUTE3 are set to NULL, and ATTRIBUTE2 contains one of the following values:

 -1 use index access if possible

 0 access path based on cost, the normal way Db2 works

 1 to nnnn Db2 should use index access on tables for which the total number of pages (NPAGES) is less than nnnn. Make sure that your Db2 Catalog statistics are up to date before you specify a value of 1 or greater.

IO Control

IO WEIGHTING in the KEYWORDS column, ATTRIBUTE2 and ATTRIBUTE3 are set to NULL, and ATTRIBUTE1 is set to DISABLE or ENABLE (deprecated in DB2 10).

SRTPOOL Modelling

SORT_POOL_SIZE in the KEYWORDS column, ATTRIBUTE1 and ATTRIBUTE3 are set to NULL, and ATTRIBUTE2 set to a positive integer up to the maximum value of SRTPOOL. That is the new SRTPOOL (for production modelling).

Production Modelling

In this case, the EXPLAIN output got changed to output which PROFILE value was active at the time of the EXPLAIN. The REASON column in the DSN_STATEMNT_TABLE gets set to “PROFILEID nnnn” for the profile number that was active at the time of the EXPLAIN.

When Was this Done?

The DSN_PROFILE_ATTRIBUTES_HISTORY has the same columns as the DSN_PROFILE_ATTRIBUTES_TABLE except that REMARKS is called STATUS and gets a value set by the START PROFILE command. Basically, a string that starts with REJECTED – or ACCEPTED – and then a text string describing why the profile was, or was not, accepted for use.

So that was it for DB2 9 – not that much but a very good start if you ask me!

System Profile Monitoring

Then in DB2 10 came “system profile monitoring”, which is where this system got very useful indeed! It then got the ability to Monitor Connections, Monitor Threads and Monitor Idle Threads.

New Keywords for Connections and Threads!

MONITOR CONNECTIONS in the KEYWORDS column, ATTRIBUTE1 is a “two part” column value. The first part is either WARNING or EXCEPTION. A warning causes a console message every five minutes depending on the diagnosis level. An exception issues the diagnosis level and rejects any new incoming connection requests. The second part is either not there or it is _DIAGLEVEL1 which issues a DSNT771I console message and _DIAGLEVEL2 which issues a DSNT772I console message with more details. ATTRIBUTE2 is a positive integer to indicate the threshold for the maximum number of remote connections. It must be less than or equal to CONDBAT. ATTRIBUTE3 is NULL. Filtering is only by the LOCATION column.

MONITOR THREADS in the KEYWORDS column, ATTRIBUTE1 is a “two part” column value. The first part is either WARNING or EXCEPTION. A warning causes a console message every five minutes depending on the diagnosis level. An exception issues the diagnosis level and can cancel the thread depending on the filtering criteria otherwise the thread is queued. The second part is either not there or it is _DIAGLEVEL1 which issues a DSNT771I console message and _DIAGLEVEL2 which issues a DSNT772I console message with more details. ATTRIBUTE2 is a positive integer to indicate the threshold for the maximum number of server threads. It must be less than or equal to MAXDBAT. ATTRIBUTE3 is NULL. Filtering on nearly all columns is allowed.

Db2 11 Docu Update

In Db2 11, an extra bit of documentation was added when filtering by Collection identifier, package name, client user name, client application name or client workstation name. When the total number of queued and suspended threads exceeds the threshold, Db2 fails subsequent SQL statements and returns SQLCODE -30041 to the client.

For example, suppose that a profile for a package is started. That profile uses ATTRIBUTE2=2. If five threads request to run the package, two threads run concurrently, two threads are queued and suspended, and Db2 fails the SQL statements for the fifth thread.

And Finally IDLE?

MONITOR IDLE THREADS in the KEYWORDS column, ATTRIBUTE1 is a “two part” column value. The first part is either WARNING or EXCEPTION. A warning causes a console message every five minutes, depending on the diagnosis level. An exception issues the diagnosis level and cancels the idle thread. The second part is either not there or it is _DIAGLEVEL1 which issues a DSNT771I console message or _DIAGLEVEL2 which issues a DSNT772I console message with more details or WARNING_MESSAGE_FOR_IDLE_TIMEOUT (only for WARNING) which issues DSNT771I and/or DSNT773I. ATTRIBUTE2 is a positive integer to indicate the threshold for the maximum number of seconds an active server thread can stay idle.

That’s all for this month, next month I will go into detail about the Filters, the new stuff In Db2 11, 12 and 13 as well as examples of different things you can do nowadays.

As always I would love to hear any comments or criticism about this topic!

TTFN

Roy Boxwell