2023-05 ZPARMs never stop changing part II

This month, I want to go through some of the absolutely most important ZPARMs that control how your Db2 systems behave in a very significant manner. All of the following ZPARMs have a performance impact of some sort. We are always trying to squeeze the last drop of performance out of our Db2 sub-systems, aren’t we?

Db2 13 and Some Db2 12 Updates Ahead!

Since this Newsletter topic first came out, in March 2022, out of the ten ZPARMs listed *five* have got new defaults! I have highlighted all these changed defaults. I have also added three new “Usual Suspects” to the list of ZPARMs that must be checked…

Starting with the Easy Stuff…

CACHEDYN. YES/NO, default YES. Should always be set to YES – unless you do not care about saving dynamic SQL performance. Back a few decades ago, the recommendation was to have this set to NO as default! Hard to believe that these days, where most shops have 80% – 90% dynamic SQL during the day!

Now we Get to the Numerics!

OUTBUFF. 400 – 400,000, default 102,400. This is *extremely* important and you really should set it to the highest possible value you can afford in real memory! As a minimum, it should be 102,400 KB (100MB). This is the buffer that Db2 uses to write log records before they are “really” written to disk. The larger the buffer, the greater the chance that, in case of a ROLLBACK, the data required is in the buffer and not on disk.

Skeletons in the Closet?

EDM_SKELETON_POOL. 5,120 – 4,194,304, default 81,920. This is one of my personal favorites, (I wrote a newsletter solely on this a few years ago). I personally recommend at least 150,000 KB and actually even more if you can back it with real memory. Just like OUTBUFF, pour your memory in here but keep an eye on paging! If Db2 starts to page, you are in serious trouble! Raising this can really help with keeping your DSC in control.

DBDs are Getting Bigger…

EDMDBDC. 5,000 – 4,194,304, default 40,960. The DBD Cache is getting more and more important as, due to UTS usage, the size of DBDs is increasing all the time.

DSC is Always Too Small!

EDMSTMTC. 5,000 – 4,194,304, default 113,386. The EDM Statement Cache (really the Dynamic Statement Cache) is where Db2 keeps a copy of the prepared statements that have been executed. So when the exact same SQL statement with the exact same set of flags and qualifiers is executed, Db2 can avoid the full prepare and just re-execute the statement. This is basically a no-brainer and should be set to at least 122,880 KB. Even up to 2TB is perfectly OK. Remember: A read from here is *much* faster than a full prepare, so you get a very quick ROI and great value for the memory invested! Keep raising the value until your flushing rates for DSC drop down to just 100’s per hour, if you can! Remember to cross check with the EDM_SKELETON_POOL ZPARM as well. It always takes two to Tango…

How Many SQLs?

MAXKEEPD. 0 – 204,800, default 5,000. The Max Kept Dyn Stmts parameter is how many prepared SQLs to keep past commit or rollback. It should be set to a minimum of 8,000 or so. Raising this might well cause a large memory demand in the ssidDBM1 address space so care must be taken.

RIDs Keep Getting Longer…

MAXRBLK. 0, 128 – 2,000,000, default 1,000,000. RID POOL SIZE is the maximum amount of memory to be available for RID Block entries. It should be at least 1,000,000 and, if you can, push it to the maximum of 2,000,000. Unless you want to switch off all RID Block access plans, in which case you set it to zero – Obviously not really recommended!

Sorts Always Need More Space

MAXSORT_IN_MEMORY. 1000 to SRTPOOL. Default 2000. The maximum in-memory sort size is the largest available space to complete ORDER BY, GROUP BY or both SQL Clauses. Remember that this is per thread, so you must have enough memory for lots of these in parallel. The number should be between 1,000 and 2,000, but whatever value you choose, it must be less than or equal to the SRTPOOL size.

Sparse or Pair-wise Access?

MXDTCACH. 0 – 512, default 20. Max data caching is the maximum size of the sparse index or pair-wise join data cache in megabytes. If you do not use sparse index, pair-wise join, or you are not a data warehouse shop, then you can leave this at its default. Otherwise, set it to be 41 MB or higher. If it is a data warehouse subsystem, then you could set this as high as 512 MB. (This ZPARM replaced the short-lived SJMXPOOL, by the way.)

Sort Node Expansion

SRTPOOL. 240 – 128,000, default 20,000. SORT POOL SIZE is the available memory that is needed for the sort pool. IFCID 96 can really help you size this parameter. Remember that the number of sort nodes leapt up from 32,000 in Db2 11 to 512,000 nodes for non-parallelism sorts and 128,000 nodes for a sort within a parallel child task in Db2 12. This means raising this ZPARM can have an even greater positive effect than before.

The Three New Guys on the Block!

To the MAX!

DSMAX used to be around 20,000 and can now be between 1 – 400,000. Remember that you will never actually reach this maximum limit as it is 31-bit memory-constrained.

Thrashing Around…

NPGTHRSH. Valid values are 0 or 1 – 2147483647. Default up to Db2 11 was 0, from Db2 12 default is now 1. SAP systems use a default of 10. The big change here, was in Db2 12 when the change from “no statistics ever ran” of -1 forced the value to be the “optimizer default” of 501 instead of the real value -1. This is also why the default is now 1 ,so that this ZPARM has a normal use! Setting it to 0 means that the access path chosen will always only be cost based.

Lock ’em Up and Throw Away the Key!

NUMLKUS. 0 – 104857600, with a default of 20,000. Just be careful raising this value too high, as each lock will take 540 bytes of storage in the IRLM!

Your “Top Ten List” + Three

These thirteen ZPARMs really influence how your Db2 system works and so must always be checked and changed with great care and attention to detail. Always do a before and after appraisal to see whether or not changing them helped or hindered your system!

If you have any comments, or other ZPARMs you think are also important for performance, feel free to drop me a line!

IDUG 2023 NA

IDUG is nearly upon again. I will be there in Philadelphia at the SEGUS booth and doing a fair bit of moderating as well. Drop on by, have a chat and pick up some of our swag and join me at the “Roy reviews AI with our WorkloadExpert” PSP on Thursday for a chance to win some cool stuff.

Hope to see some of you there!

TTFN,

Roy Boxwell

2021-05 Soundex and other cool features part 7 for Db2 12

In this, obviously, never ending series of new features, I will roll up all the new ones since my “SOUNDEX and other „cool“ features – Part six All new for Db2 12” newsletter from 2018-08 where I first covered the new stuff in Db2 12.

What was new in FL100

At that level, a new table function called BLOCKING_THREADS was introduced, which is primarily used by the command DISPLAY BLOCKERS to help in looking for bad guys who are ready to BLOCK your change to FLxxx

FL500 Changed Scalars

Two scalar BiF got changed: TRANSLATE and UPPER can now both use UNI_60 as a locale.

FL500 New Scalars

In FL500, we got a nice bunch of new Scalar functions:

GENERATE_UNIQUE_BINARY (like GENERATE_UNIQUE except it returns a BINARY(16) value)
HASH_CRC32
HASH_MD5
HASH_SHA1
HASH_SHA256
PERCENTILE_CONT
PERCENTILE_DISC
WRAP

FL500 Changed Aggregates

One aggregate BiF got updated: ARRAY_AGG can now be used for associative arrays.

FL501?

It brought in just one new Aggregate BiF, which was the very famous LISTAGG. This is famous, as it was the “proof of concept” that Agile actually worked and could function in the wild.

FL502?

Two changes to Scalar BiFs in this release:

GRAPHIC The first argument now accepts numeric data types, including SMALLINT, INTEGER, BIGINT, DECIMAL, REAL, DOUBLE, FLOAT, and DECFLOAT.
VARGRAPHIC The first argument now accepts numeric data types, including SMALLINT, INTEGER, BIGINT, DECIMAL, REAL, DOUBLE, FLOAT, and DECFLOAT.

FL503?

Nothing new for us in that release!

FL504?

It brought in a brand new concept for Db2 on z/OS. The concept of “pass-through BiFs”. These are just checked for valid syntax before being passed on down to the Accelerator that would actually execute them. Obviously, you must be careful about their usage!

CUME_DIST
CUME_DIST (aggregate)
FIRST_VALUE
LAG
LAST_VALUE
LEAD
NTH_VALUE
NTILE
PERCENT_RANK
PERCENT_RANK (aggregate)
RATIO_TO_REPORT
REGEXP_COUNT
REGEXP_INSTR
REGEXP_LIKE
REGEXP_REPLACE
REGEXP_SUBSTR

FL505?

Brought in a whole bunch of encryption and decryption BiFs:

ENCRYPT_DATAKEY converts a block of plain text to a block of encrypted text using a specified algorithm and key label.
DECRYPT_DATAKEY_datatype functions return the block of encrypted text from ENCRYPT_DATAKEY as a block of the datatype specified by the function used. List of datatypes: DECRYPT_DATAKEY_INTEGER, DECRYPT_DATAKEY_BIGINT, DECRYPT_DATAKEY_DECIMAL, DECRYPT_DATAKEY_VARCHAR, DECRYPT_DATAKEY_CLOB, DECRYPT_DATAKEY_VARGRAPHIC, DECRYPT_DATAKEY_DBCLOB, and DECRYPT_DATAKEY_BIT.

FL506?

Here IBM came up with a slew of “alternative names” for existing BiFs and a new method of calling HASH routines. The background here was to make porting to Db2 on z/OS that little bit easier!

CHAR_LENGTH CHARACTER_LENGTH, which returns the length of its argument in the number of string units that are specified
COVAR_POP COVARIANCE or COVAR, which return the population covariance of a set of number pairs
HASH HASH_MD5, HASH_SHA1, or HASH_SHA256, which return the result of applying a hash algorithm to an input argument, depending on the value specified for the second argument for the HASH function:

        0 (default) HASH_MD5 Returns VARBINARY(16) instead of BINARY(16)
        1 HASH_SHA1          Returns VARBINARY(20) instead of BINARY(20)
        2 HASH_SHA256        Returns VARBINARY(32) instead of BINARY(32)

POW POWER, which returns the value of one argument raised to the power of a second argument
RANDOM RAND, which returns a double precision floating-point random number
STRLEFT LEFT, which returns a string that consists of the specified number of leftmost bytes or the specified string units
STRPOS POSSTR, which returns the position of the first occurrence of an argument within another argument
STRRIGHT RIGHT, which returns a string that consists of the specified number of rightmost bytes or specified string units
TO_CLOB CLOB, which returns a CLOB representation of the first argument
TO_TIMESTAMP TIMESTAMP_FORMAT, which returns a timestamp for a character string expression, using a specified format to interpret the string

FL507?

Here IBM development added another slew of “pass through” BiFs:

ADD_DAYS
BTRIM
DAYS_BETWEEN
NEXT_MONTH
REGR_AVGX
REGR_AVGY
REGR_COUNT
REGR_INTERCEPT or REGR_ICPT
REGR_R2
REGR_SLOPE
REGR_SXX
REGR_SXY
REGR_SYY
ROUND_TIMESTAMP if invoked with a date expression

FL508?

None that I have read of yet!

FL509?

None yet…

Naturally I will be keeping this newsletter up-to-date, as necessary.

Any questions or ideas, do not hesitate to drop me a line,

TTFN,

Roy Boxwell

2020-04 Four Flavors of Db2 Audit

As well as simply getting it done, I will also run through the four different ways that you can process the gathered data.


Four ways to get a Db2 z/OS Audit done


1- First up

First option is the simplest, cheapest and quickest:

Do nothing.

Whether or not this will help your company is a non-trivial question of course!

Naturally this is an absolute No No.


2- Then we have

Next option is relatively simple and cheap, but requires a bit of work: 

Write it all yourself but based on existing data that some other process already extracts for you, (SMF for example). 

If you happen to have the skills for extracting the required audit data from existing data that is being collected anyway, then this might well be the best method if you are really strapped for resources. 


3- Getting there 

Then we have not so simple, still cheap, but a ton of work: 

Write it all yourself and add all the IFCIDs you actually need to audit your system as well as capturing all the SQL. 

This needs a serious amount of skills to get and keep up with the agile world of Db2. You will also need to take care of the amount of data that you will be collecting.

However, the auditor will be happy as you have everything they could ask for.


4- Aha! The only true way 

Last option is simple, not so cheap but very quick: 

Third party software that does it all for you.

This is my preferred solution, especially as we just happen to sell one (WorkLoadExpert Audit).

This is actually the only real way to go. You probably don’t have the time to keep all these things up-to-date and running correctly. 

Data Collected – Now what? 

So, you have chosen one of these ways to gather the data. Now you must evaluate what you got. Here again we have four separate ways to go forward: 

First up 

There it is! 

Do nothing. Just point at the datasets, print outs, database objects and say “It is all in there…” 

This is not really a solution and any auditor worth his, or her, salt would quite rightly be extremely upset! 

Then we have 

A whole bunch of pre-written SQLs. 

SPUFI is ok, but much better would be to see these in a GUI where graphical viewing is built in and saving and sharing results is much easier.  

This is not bad, but still a manual “island” process. Just Db2 and nothing else plus it must be triggered by humans. 

Getting there

A whole bunch of pre-written and custom SQLs.

This time, all run in Batch and the results are emailed to the auditor directly. These emails can “just sit there” until the auditor checks the results. Naturally, if anything is found, then the underlying data must still be there for a detailed analysis.

Better, as it is getting automatic but still not really “round”, as it is still Db2 in isolation…

Aha! The only true way

Use of LEEF or SYSLOGGER-style formats to export all audit data.

The data is then in a data-lake where SPLUNK, QRADAR et al can happily slice and dice their way through the data.

This is the best way!

You also get an extra bonus point for *removing* the data from the mainframe. As auditors *love* a single point of control, this is the only real way forward. It also pushes the Db2 data into the world of other data that auditors use and require.


Db2 Audit with “GIVE&TAKE” :


Software Engineering GmbH and SEGUS Inc are launching a new free Give&Take which this time is the Audit support from WorkLoadExpert.

If you would like to take part, then please just fire off an email to db2support@segus.com telling us who you are and which firm you work for and we will get in touch!

Give and Take 

By the way, it is called “Give&Take” because :

  • we Give you the software, for free, to run for a trial period, and
  • we would like to Take away what you think, feel, and find about the software after the trial period. 

More about Give&Take


TTFN, 

Roy Boxwell 

2019-12 Fun with Db2 12 PBR RPN

I have recently enhanced our SpaceAssuranceExpert (SAX) product to automatically help out if partitions, or partitioned indexes, are getting too big for their boots in a productive system.

SAX – What is it?

The idea behind SAX, is to stop any and all chances of getting a dreaded SQLCODE -904 in production, especially “out of the blue”.

Our idea was, in Db2 12, with the new PBR RPN to do an on-the-fly ALTER to the DSSIZE, just like SAX does now with the SECQTY to avoid running out of extents.

RTFM Time

A quick look in the manual tells you that :


it is an immediate ALTER (as long as you make the new DSSIZE larger than the old one!) and there are no package invalidations or REORGS required.


This is fantastic!

So I created a nice little PBR and then ran a horrible Cartesian join SPUFI to flood the first partition with data. This join had a TIME card of (,1) to limit it to one second of CPU before getting an Abend S322.

SAX Monitor

The SAX monitor reacted perfectly and did the TP alter to 513 GB DSSIZE (I had an increment size of 512 as a test), and got an SQLCODE -666 (I just love that SQLCODE…), as the INSERT was still running.

This ALTER was then internally queued to be attempted later, like in any other failure case.


All well and good.

Wham! Nasty errors !

Then I did the same for a DPSI on my PBR RPN… Oh dear!I got a nasty IO Error and then an even nastier ROLLBACK loop, meaning I had to cancel the IRLM to stop Db2… (There is an APAR for this problem PH18977.)

Rollback loop?

The ROLLBACK loop was caused by me choosing to use NOT LOGGED as a tablespace attribute. We have a finite amount of log space, and when the transaction was S322’d after one second of CPU,

the rollback could not find one of the archive logs and then we had to cold start Db2

– Not pretty!

Proper test!

I then wrote a couple of little test programs that actually COMMITted after 5000 inserts and then the ALTERs all worked as designed.


Do I worry too much about extended format and extended accessibility?

Next, I worried about the ominous “extended format and extended addressability“ attributes in the DATACLASS for a PBR RPN, and wondered what would happen if a customer has SAX running and it happily ALTERs a TP to say 6 GB when they can only address 4 GB…


IBM to the rescue!

Luckily for us, Db2 development had thought about this!

  • If you attempt to create a PBR RPN (even with a very small size) and your DATACLASS does not have the two attributes set, you get an error message 00D70008 telling you this detail.

  • If you ALTER an existing tablespace to be a PBR RPN and your DATACLASS does not have the two attributes set, then the ALTER works fine. But remember, this is a pending alter and you *must* do a REORG at the TS level with inline TP level copies. This REORG then fails – also with 00D70008.

So, in other words, SAX cannot hurt you here!


What about PBGs?

After all this we also considered PBGs. They have a limit as well – MAXPARTITIONS in their case. So we added an ability to also ALTER MAXPARTITIONS as well. Here you must be more careful though, as these ALTERs are still immediate *but* they invalidate any referring packages!

ABIND YES or NO?

If you work with ABIND set to YES all is good as the ALTER comes in, Db2 invalidates your packages and the auto rebind happens so fast that you do not even notice it happening, however, if you work with ABIND NO then any packages, even the package actually running and doing the inserts, will fail! If you are just using dynamic SQL then it is 100% OK otherwise – Buyer beware!

Aha!

I have opened an Analytics Idea (DB24ZOS-I-1057) to try and get this loophole closed, as I cannot see what access path change could be affected by going from 32 to, say, 36 MAXPARTITIONS.

Db2 keeps adding the parts dynamically and nothing happens then… Go figure… Anyway, if you would like it – Please vote for it!


There are already some nice comments attached to it:

  • DP commented

    this is just a limit in the catalog.  So how could access path be affected?  The actual growth of a partition doesn’t invalidate the package so how just changing the limit do so ?
  • BW commented

    I opened a Case on this asking why packages are being invalidated in this situation and it is still open waiting for a reply.
  • BD commented

    With only impact to catalog, not sure how Access Path would be impacted.  Seems wasteful and counter productive to invalidate packages.

Remember – You never stop learning!

As always I would be pleased to hear from you!

TTFN,
Roy Boxwell
Senior Architect

2019-08 FTB (Fast Traversal Block): Just another TLA?

So we get a new TLA (Three Letter Abbreviation) to try and wrap our heads around in the busy busy world of mainframe databases. This time it comes with a twist, as FTB is actually not “Feed The Beast” sadly, but “Fast Traversal Block”.

In reality, it is referred to everywhere as:

FIT (Fast Index Traversal)

All clear on this and still with me? Good then I may begin!

The problem with modern indexes

The core problem these days is time… the time it takes to go through the leaf pages down to the actual data page is just too long. The vast amount of data we have forces upon the index a large number of layers which, in turn, forces a lot of non-leaf page accesses.

Index structure

We all know how indexes look in Db2, but remember they all have a root page (which is always technically speaking a non-leaf page) which points, normally, to non-leaf pages which, in turn, point to either more non-leaf pages – so you are getting extra levels in the index here  – and finally they point to a leaf page which actually contains pairs of keys and RIDs so that Db2 can get the data from the table page.

Which level are you?

One of the quirks of Db2, is that the leaf pages are called Level 0 and then the non-leaf pages directly above them are Level 1. The root page above these is then Level 2 and so you have a “two level” index.

So how many fetches for a random data access?

The first thing Db2 does is read the root page, looking at the root page it sees which non-leaf page must be fetched and it repeats this until it lands on the leaf page and then it fetches the data. So in my two level index it takes four fetches to get to the data. Now imagine you have a seven level index? Yep you end up fetching a ton of non-leafs!

The solution? FTB!

In Db2 12, IBM introduced the Fast Traversal Block to keep a copy of the non-leaf data in a new area, separate from the buffer pools.

The major reasons for the expected performance boost, are the facts that the structure is L2 cache-aware and each page is equal in size to one cache line (256 bytes), and I/O is massively reduced for the “other” non-leaf pages.

One size fits all?

The size of the FTB is determined by the new ZPARM INDEX_MEMORY_CONTROL with valid values AUTO, DISABLE or 10 – 200,000 MBs. By default AUTO is on. With this value, the size of the FTB is either 20% the size of all available buffer pools or 10MB, whichever is the highest.

The devil is in the detail

To actually start working with FTBs, you have to wait until the FTB Daemon has found an eligible index. From all of your indexes there will be quite a few candidate indexes. These must follow the following rules:


  1. It cannot be longer than 64 bytes

  2. It cannot be versioned (So the OLDEST_VERSION and CURRENT_VERSION must be the same!)

  3. It cannot contain a TIMESTAMP column with TIMEZONE

  4. It cannot have more than 2,000,000 leaf pages


Once through that selection list, the Daemon starts seeing what is happening to this index, is it a random select? That’s is a good thing! Is it an index only access? That’s even better! A split! Oh dear, that’s very bad…from all this, every two minutes, the daemon decides whether or not to use FTB for this index.

Control is in your hands

This is all well and good, but there are a lot of people who like to see *exactly* what is going on and where! So IBM introduced a new command:

-DIS STATS(IMU)

From this you get to see which indexes are being processed and how much space they are using. The Daemon also spits out console messages so you can see what happened and why. Further there is a new Catalog table SYSIBM.SYSINDEXCONTROL where you can micro-manage exactly which indexes to attempt to force into FTB and which ones to exclude. I would try and avoid this option and just let Db2 do the business!

Gotcha’s?

Always! FTB is limited to 10,000 per member which is probably ok! When a MASS DELETE (or TRUNCATE TABLE) happens then the FTB is removed for the related indexes and, even more importantly, is to keep up with current Db2 maintenance! As always with new things, it takes a while to get the birthing pains out of the way!

INCLUDE can kill you!

Finally, remember that the length cannot be greater than 64 bytes. So if you decide to add that one byte character column to the index to get Index Only access, you might then push the index over the edge and up to 65 bytes…

Before ALTERing indexes always check if they are in the FTB (or were!) and check if your alteration will disallow the FTB!

A little SQL to show you what you have

WITH INPUT (NLEVELS, LENGTH, INDEX_NAME) AS                            
 (SELECT  COALESCE(E.NLEVELS , A.NLEVELS )  -- FIRST RTS THEN INDEXES   
       , SUM(CASE D.COLTYPE                                             
             WHEN 'DECIMAL ' THEN                                       
 -- IF , SEPERATOR           SMALLINT( CEILING(( D.LENGTH + 1,0 ) / 2 ))
                             SMALLINT( CEILING(( D.LENGTH + 1.0 ) / 2 ))
             WHEN 'GRAPHIC'  THEN D.LENGTH * 2                          
             WHEN 'VARG'     THEN D.LENGTH * 2                          
             WHEN 'LONGVARG' THEN D.LENGTH * 2                          
             ELSE D.LENGTH                                              
             END)                                                       
       + SUM(CASE A.PADDED                                              
             WHEN 'Y' THEN 0                                            
             ELSE                                                       
               CASE D.COLTYPE                                           
               WHEN 'VARG'     THEN 2                                   
               WHEN 'LONGVARG' THEN 2                                   
               WHEN 'VARCHAR'  THEN 2                                   
               WHEN 'LONGVAR'  THEN 2                                   
               WHEN 'VARBIN'   THEN 2                                   
               WHEN 'DECFLOAT' THEN 2                                   
               ELSE 0                                                   
               END                                                      
             END)                                                       
       + SUM(CASE D.NULLS                                               
             WHEN 'Y' THEN 1                                            
             ELSE 0    
                END) AS LENGTH  
       , STRIP(A.NAME) CONCAT '.' CONCAT STRIP(A.CREATOR) AS INDEX 
 FROM SYSIBM.SYSINDEXES        A                                       
     ,SYSIBM.SYSKEYS            C                                       
     ,SYSIBM.SYSCOLUMNS         D                                       
     ,SYSIBM.SYSINDEXSPACESTATS E                                       
 WHERE A.UNIQUERULE     <> 'D'              -- NOT DUPLICATE            
   AND D.COLTYPE        <> 'TIMESTZ'        -- NOT TIMEZONE             
   AND A.OLDEST_VERSION = A.CURRENT_VERSION -- NOT VERSIONED            
   AND A.TBNAME         = D.TBNAME                                      
   AND A.TBCREATOR      = D.TBCREATOR                                   
   AND A.NAME           = C.IXNAME                                      
   AND A.CREATOR        = C.IXCREATOR                                   
   AND A.NAME           = E.NAME                                        
   AND A.CREATOR        = E.CREATOR                                     
   AND C.COLNAME        = D.NAME                                        
 GROUP BY A.NAME, A.CREATOR , E.NLEVELS, A.NLEVELS)                     
 SELECT NLEVELS, LENGTH , INDEX_NAME                                    
 FROM INPUT                                                             
 WHERE LENGTH <= 64   
 ORDER BY NLEVELS DESC
 FOR FETCH ONLY       
 WITH UR              
 ;                                                    

I have limited it to just show you just the eligible unique, non timezone, non-versioned indexes with a length of 64 or less of course!

Blog time


Here’s an excellent write up from John Campbell with even more technical details:

https://www.ibm.com/developerworks/community/blogs/897a7c98-57af-4523-9cfa-07ebc3f996b4/entry/Db2_12_greatest_hits_with_John_Campbell_fast_index_traversal?lang=en


and from Akiko Hoshikawa: a very good IDUG Blog:

https://www.idug.org/p/bl/et/blogaid=646


As always I would be pleased to hear from you!

TTFN,
Roy Boxwell
Senior Architect

2019-06 A little quote from Terry Purcell: it just takes one

This is a little quote from Terry Purcell which I think should be understood much better than it possibly is today!

The quote is based on the idea that :

One single SQL statement can bring your entire system to its knees.

Really? Just one?

Many people think this is a huge exaggeration – However it is not!

It happened to me

This is not a friend of a friend story, but a story that really happened. Names have, of course, been changed but the story itself is 100% accurate.


The story


It was a normal Db2 PTF Maintenance night…

At this shop they always put maintenance in at midnight on Saturdays. All done automatically, and then roll the update around the data-sharing group so that, relatively quickly, all sub-systems have the same PTF level.

So far so normal

Sunday is not that busy at the site and no-one noticed any change.

Monday was different

Monday morning the machine started acting up. Customers could not login and the help desk was 100% busy. The lead DBA, George, had by around 08:00 a.m. received the first escalation to management and so the number of calls went up.

The bad guy

Using our WorkloadExpert (WLX) software together with our Bind ImpactExpert (BIX) the other DBAs, Fred and Ringo, quickly identified one single SQL statement that was taking 30,000 times more resources than in the prior week!

This was an SQL with table functions and LEFT, INNER JOINs etc.

What was it?

The comparison in BIX showed that all that had changed was one index access was now matchcols one instead of matchcols two! That was it! This one tiny change on one little SQL killed this machine…

The fix?

They decided to roll back the PTFs and quickly did this and everything returned to normal…


The lesson?

Always test any PTF before you go live – as just one SQL can kill ya!

Terry had indeed warned us all!


How can we help?

Well, thank you for asking: Our product CDDC contains an SQL replay and compare function that would have spotted this SQL in two different complementary ways. Firstly the BIX part of CDDC would have spotted it straightaway and secondly the replay itself would have thrown this out as a major outlier and bad guy candidate!

Different ways to Rome

They could possibly even have fixed this without backing out the PTF apply. They could have tried using our RUNSTATS Rescue to attempt to use older statistics and see if one of them would have given them a matchcols two access path. This time, however, with all the managers breathing down their necks it was decided – Undo all the changes!


In your shop

How would you have handled this situation in your shop?

Would/Could you have seen this before it happened?

Feel free to send me your comments and ask questions.

TTFN,
Roy Boxwell
Senior Architect

Machine Learning for Db2 z/OS: Artificial Intelligence – Hype or Reality?

Now that the Db2 Optimizer has gone all Artificial Intelligent (AI from now on) on us, I thought it would be interesting to review where we are in this brave new world!

Keeps you out of a BIND

My firm created our BindImpactExpert (BIX) software many moons ago to compare access paths, typically before a program non SQL change or a REBIND after RUNSTATS etc. Just to warn you not to do the REBIND or whatever you were going to do, as it would change the access plan to something you may well not really want to have. This means not just no-brainers like index access to tablespace scan, but also different index usage or table access order changes.

Apples and Oranges

So, what has that got to do with AI?
Well, we used a marketing slogan of trying to compare oranges and apples to try and make it clear :

how tough it really is to try and match SQL statements across different program releases and to compare the results.

ML and AI – The new buzzwords

Now in Db2 12, the optimizer has got machine learning (ML from now on) to help it study which host variables are used the most and how many rows are usually returned by a given query.

Do you “know” your data?

When the optimizer “knows” this info it can of course make much better “guesses” at good repeatable access plans that help everyone! Now the usage of AI and ML always gets me interested because of all the fun you can have with them.


There was recently a great article in my favourite science magazine “New Scientist” all about how to fool image recognition AI:

https://www.newscientist.com/article/mg24232270-200-machine-mind-hack-the-new-threat-that-could-scupper-the-ai-revolution/


Now you should see the connection back to our BindImpactExpert slogan. So much for Oranges and Apples – AI cannot even see that an Orange is an Orange…

this is not good or even perhaps “suboptimal”…

Adversarial Images and the fun you can have with them…


From 2018 this article

https://www.wired.co.uk/article/artificial-intelligence-hacking-machine-learning-adversarial

Within this article is a great paper all about graffiti changed road signs! Here is a direct link to the article https://arxiv.org/pdf/1707.08945.pdf


Fooling Deep Learning

Another great paper where Deep Learning AIs jump off the deep end:

http://www.evolvingai.org/fooling


Disappearing Humans

and finally this joyous article and a link to a You Tube video:

https://www.theverge.com/2019/4/23/18512472/fool-ai-surveillance-adversarial-example-yolov2-person-detection


But do not wear this picture as a T-Shirt down the street anytime soon!!!

Would you sit in a driverless car?

What this all boils down to is that I, for sure, will never sit in a driverless car within the next 40 years!
Why?

  • Firstly, I believe a true AI does not exist yet and will not exist for at least another 40 years.
  • Secondly, when I see an Orange I see an Orange because I “know” what an Orange looks like! As you have seen AIs have no concept of what an Orange *is* and so can easily see a drill instead or mistake skiers for dogs…
  • Thirdly, I am a programmer and someone somewhere is writing and testing – I hope! – the code… need I say more? Don’t even get me started with Agile or KANBAN.

Artificial Intelligence - Machine Learning for Db2 z/OS
extract from © 2019 – „We need to talk, AI” – Dr. Julia Schneider und Lena Kadriye Ziyal
A Comic Essay on Artificial Intelligence

https://weneedtotalk.ai


Room full of CTO’s

Have you heard about the hopefully apocryphal story of a room full of CTO’s at a software conference when they were asked “Would you fly in a computer controlled aircraft?” All their hands went up. Then they were asked “Would you fly if it was your software teams that had written the flight control software?” all their hands went down!

Flying is strangely different

While talking about flying my answer would be different?

Why?
Because in the air you have no-one else apart from other highly automated devices anywhere near you! This is the one hope that I can see for the whole driverless car stuff – Driving on the Interstate, Motorway or Autobahn. In the air, the environment is much more rigid. No lights, no crossings, no opposite traffic, no foot traffic, no children chasing a ball across the road etc.

They will come…

I am sure that driverless cars will come soon and I am equally sure that people will die because of them – However, I am also sure that lots of accidents will be avoided as well. After all the AI will always react faster than a human.

With this comes the ethics of AI and ML of course which will also play a major role. If the car has to decide in a crash scenario :

  • who gets hurt, should it protect itself and its cargo above all others?
  • What about a group of playing children or a bus full of nuns?
  • How can it decide who lives and who dies?

Looking on the bright side at least it will not get drunk, or high, and drive! I would also seriously hope that AI does not get road rage!

But even with these “bonus” points, I still have massive doubts and serious worries about the quality of the software and the true image recognition potential of current AIs.

AI everywhere

We were recently asked by a customer to add AI and ML to our SpaceAssuranceExpert (SAX) software. This has been happily running for years, capturing dataset allocations and extent usage in Db2 and making sure they never ran out of space or extents without getting told about it well beforehand.

I had to do a real double take and think “How on earth could AI help here?”

You are approaching 32 Linear Datasets with your segmented tablespace. You have to ALTER and REORG before you use up all the space in the 32nd. Where can AI help here?
We already have extrapolation from current usage…

Don’t get me wrong

I am not against AI and ML…

but I think we *all* need to get a bit more skeptical about what can actually be delivered now.

On the one hand host variable contents and row counts – Yes indeed, but even here outliers will cause grief and then on the other hand the complete and utter science fiction of self-driving cars in our towns and cities.

The money being spent is also amazing, in 2019 it is estimated, by Syncsort,

that the AI budget will be $35.8 Billion 44% more than in 2018.


What do you think?


I am sure that autonomous cars will be a plague upon us – are you?

TTFN,
Roy Boxwell
Senior Architect

PS: I am looking forward to the first Adversarial Images for Db2 z/OS !!!

2019-04 ZOWE for Continuous Delivery – It’s worse than that – he’s tested Jim!

Terrible quote usage I know… But now, with the all new and Agile Db2 12 being picked up and used globally, I think it is time to review a few things that companies and individuals must do, or at least plan, in this “Brave New World” of ours.

To wrap up at the end, I will delve into the idea of using ZOWE to help you test!


This Db2 12 testing checklist helps to navigate into the Db2 Continuous Delivery world, well supported by Zowe, the modern IBM z/OS GUI

Table of contents – Db2 12-testing-checklist
  1. Cloning
  2. Workload collection
  3. Dynamic & Static SQL storage
  4. IFCID376 & co
  5. EXPLAIN-& Access Path
  6. Execute the SQLs
  7. Db2-12 Test Review
  8. Zowe – Open Source z GUI – for CD

The Db2 12 Testing Checklist


1. Do not be afraid of the clones!

To even begin to start testing Db2 12, you’ll need to start with a complete production clone. I know a lot of people get suddenly scared about this due to audit requirements or, more commonly, space requirements. I mean, who can possibly support a complete 100% data clone of their productive system?

A real clone or a partial clone?

Of course you do not actually need all of your data! All you must have is the Db2 Catalog and Directory, all of the user objects should be DEFINE NO style empty shells. After all, do you really need TBs of productive data to do a test run? Nope! So now we have removed two major problems (Space and Audit) from the list of testing problems.


2. Collect as much workload as you can

You’ll need to gather as much of your current executed SQL as you possibly can. At best 13 months is a good target. “Not possible” I hear you all scream – “Oh yes it is!” say I.


3. Dynamic & Static SQL: Store it cleverly away

You do not need the same SQL statement 12,000,000 times. Just once, but with an execution count. Store the dynamic SQLs and the retrieved Static SQLs away somewhere nice and safe with a COMPRESS YES to save room, and then you can easily get 13 months of data. Another major problem gone from the list of testing problems.


4. Gather your friends close but your enemies closer – IFCID376

Make sure you are running with IFCID 376 always on and everywhere! Test, QA and Prod. The moment it starts spitting out possible “problems”, start trying to track down the causing event (SQL) and verify that it is OK, or get a code change implemented. This only has a little to do with agile Db2 12 but should be on at every shop anyway!

But I don’t get the Dynamic SQL!

You do if you are fast and good enough! If you are permanently snapping the DSC and catching the flushed statements, then you can indeed find out the SQL that caused the IFCID. Another major problem disappears from the list of testing problems.


5. Access Paths a go go

Now it really makes a lot of sense to EXPLAIN all of the SQL that you have before and after the Db2 12 Functional Level (FL) change to see if just flipping to, for example, FL 504 caused access paths to head south… This is something you should all be doing now as well of course. Always validate the access path before being forced to do a REBIND or PREPARE. This way you will not get any nasty surprises… Another major problem vanishes from the list of testing problems.


6. Execute the SQLs

I have one last thing for you…

Take all of the SQL that you have just processed and transform it so it can actually execute.

  1. Logarithmically reduce the execution count and execute all of the SQL while monitoring all of the Db2 KPIs – On the Clone, of course.
  2. Then reset the world back to your starting Clone, do the FL change and then re-execute all the SQL again.
  3. Once finished, compare all of the KPIs – looking for outliers as there has only been the FL change then.

Any differences you see should be just “background noise” but there could also be some nasties in there. This will light them up so that the root problem(s) can be found and corrected way before you actually do the FL change in production.


Well done, you’ve made it!

That’s it! With all these processes set up and fully automated you can actually do repetitive pipeline tests within hours instead of weeks or months or never!


7 – Db2 12 test-Review:


What does this mean for me?

Well,…

  • You need to get a fast, good, automated Cloning system in place.
  • You need to get a fast, good, automated IFCID 376 system in place.
  • You need to get a fast, good, automated SQL workload collection system in place.
  • You need to get a fast, good, automated Access Path comparison system in place.
  • You need to get a fast, good, automated SQL replay system in place.

… this means :

” Continuous testing in a continuous development world “…


You might, in fact, need to actually buy some software that does all this for you…

May I introduce you to our newest member of the SEGUS/SEG Family:

Db2 Continuous Delivery Deployment Check

is the software that does all of this, and more!

Check out our website for details of how this software really helps in :



  • Reducing the time to test

  • Enabling you to actually go forward with a good, clear conscience!

Visit our CDDC

CDDC for Db2 z/OS - Continuous Delivery Deployment Check - Agile & Environment simulation, Zowe IBM mainframe GUI

8 – ZOWE – Open Source zUI – to the rescue?

You have probably heard of ZOWE by now, the first open source software on z/OS from IBM, Rocket Software and CA Technologies (A Broadcom company). It was launched last year and is growing very quickly.

Introduction to Zowe:

https://www.youtube.com/watch?v=NX20ZMRoTtk

My firm is also using it now and all future developments will also be ZOWE enabled.

In a nutshell,

it allows users to interact with the Mainframe using a modern GUI.

This is based upon z/OSMF and uses work flows to actually do stuff. You define a “micro service” to do one thing and then you can string as many of these together as you like to get something done like, e.g. Provisioning a Db2 system or Cloning a Db2 subsystem etc.


The IBM web based UI for Db2 z/OS


Zowe at SEGUS and

SOFTWARE ENGINEERING?



We will be using ZOWE for two things:

1 – For the Installation and Maintenance of our software at customer sites

ZOWE enables people to do things that normally require “green screen” and we are keenly aware that green screen people are disappearing. If we want the z/OS platform to survive, it must be dragged, kicking and screaming, into the modern world.

2 – For using ZOWE as the front end for the users of our products

The CDDC product that I described in this newsletter will be built around ZOWE, thus expanding its usability in the market of tomorrow.


We are convinced that ZOWE is the way to go – are you?


TTFN,
Roy Boxwell
Senior Architect

2019-03 EXPLAIN data review

A quick history of EXPLAIN and a new standard Db2 EXPLAIN Query to bring out the best and most underused columns.

EXPLAIN has been with us from nearly the start of Db2 (DB2 V2R1 when the b was B!) and, over the years, the number of Explain tables has mushroomed up to 20 in Db2 12. Not *all* of this data is really useful but there are bits and pieces that are well worth adding to your standard repertoire of explain queries!

Table of contents
  1. PLAN_TABLE
  2. DSN_DETCOST_TABLE
  3. DSN_FILTER_TABLE
  4. DSN_PREDICAT_TABLE
  5. New Standard Explain SQL

PLAN_TABLE

This can be used to see in which release the PLAN_TABLE was created:

V2.1 – 25 column format
V2.2 – 28 column format
V2.3 – 30 column format
V3.1 – 34 column format
V4.1 – 43 column format
V5.1 – 46 column format
V6.1 – 49 column format
V7.1 – 51 column format
V8.1 – 58 column format
V9.1 – 59 column format
10 – 64 column format ending with MERGN
11 – 66 Column format ending with EXPANSION_REASON
12 – 67 column format ending with PER_STMT_ID

What could be interesting here?? Well what about:

PREFETCH                           D, S, L, U, or Blank?

  • D for Dynamic
  • S for pure Sequential
  • L for through a page list
  • U for unsorted RID list
  • Blank for Unknown or no prefetch

COLUMN_FN_EVAL R, S, X, Y, or Blank?

  • R for while data is read
  • S for while performing a sort
  • X for while data is read but using OFFSET
  • Y for while performing a sort but using OFFSET
  • Blank for after data retrieval and any sort

PAGE_RANGE Y or Blank.

  • Y for yes the table qualifies for page range screening
    so that only the needed partitions are scanned
  • Blank for no

PRIMARY_ACCESSTYPE D, P, S, T, Blank. Is direct row access attempted first:

  • D it tries to use direct row access with a rowid column. If it cannot do this it uses the access path that is described in the ACCESSTYPE column
  • P it uses a DPSI and a part-level operation to access the data
  • S it uses sparse index access for a sideways table reference
  • T the base table or result file is materialized into a work file, and the work file is accessed via sparse index access. If a base table is involved, then ACCESSTYPE indicates how the base table is accessed
  • Blank it does not try to use direct row access by using a rowid column or sparse index access for a work file.

DSN_DETCOST_TABLE

There are a ton of interesting columns here but most are “IBM internal only”, however, these are available for our enjoyment:

ONECOMPROWSThe number of rows qualified after applying
local predicates.
IMLEAFThe number of index leaf pages scanned
by Data Manager.
IMFF  The filter factor of matching predicates only.
IMFFADJ  The filter factor of matching and screening
DMROWS  The number of data manager rows returned
(after all stage 1 predicates are applied).
RDSROWThe number of data manager rows returned
(after all stage 1 predicates are applied).
IXSCAN_SKIP_DUPSWhether duplicate index key values
are skipped during an index scan.
  • Y Duplicate key values are skipped
  • N Duplicate key values are not skipped
IXCAN_SKIP_SCREENWhether key ranges that are disqualified
by index screening predicates are
skipped during an index scan.
  • Y Disqualified key ranges are skipped
  • N Key ranges are not skipped
EARLY_OUTWhether fetching from the table stops
after the first qualified row.
  • Y Internal fetching stops after the first
    qualified row
  • N Internal fetching continues after the first
    qualified row
BLOCK_FETCH or N Was block fetch used?

DSN_FILTER_TABLE

Sometimes it is really interesting to see when the predicate is applied and whether or not it could be pushed down.

STAGEThe stage that the predicate was evaluated.
MATCHING, SCREENING, PAGERANGE,
STAGE1, or STAGE2.
PUSHDOWNWhether the predicate was pushed down.
  • I for the Index Manager evaluates it
  • D for the Data Manager evaluates it
  • Blank means no push down was used

DSN_PREDICAT_TABLE

Here lives the really good stuff that most people do not use! The Bubble Up is normally an SQL coding error and the T is “forgetfulness”…

ADDED_PREDIf this column is non-blank it means that Db2 has
rewritten the query to some extent.
This is not good as it adds CPU to the process.
In my opinion any “added predicate”
should already be coded correctly in the SQL!
  • Blank Db2 did not add the predicate
  • B for bubble up
  • C for correlation
  • J for join
  • K for LIKE for expression-based index
  • L for localization
  • P for push down
  • R for page range
  • S for simplification
  • T for transitive closure

CLAUSE

Is this a SELECT, HAVING, ON, or WHERE clause?
ORIGIN  Where did it come from?
  • C for a column mask
  • R for a Row Permission
  • U specified by User
  • Blank generated by Db2

New Standard Explain SQL

Just adding these to your standard Explain query like this:

SET CURRENT SQLID = 'BOXWELL' ;                                   
SELECT SUBSTR(P.PROGNAME, 1 , 8 ) AS PROGNAME
,SUBSTR(DIGITS(P.QUERYNO), 6, 5) AS LINE
,SUBSTR(DIGITS(P.QBLOCKNO), 4, 2) AS QNO
,SUBSTR(DIGITS(P.PLANNO), 4, 2) AS PNO
,SUBSTR(DIGITS(P.MIXOPSEQ), 4, 2) AS SQ
,SUBSTR(DIGITS(P.METHOD), 5, 1) AS M
,SUBSTR(P.TNAME, 1, 18) AS TABLE_NAME
,P.ACCESSTYPE AS A
,P.PRIMARY_ACCESSTYPE AS PA
,SUBSTR(DIGITS(P.MATCHCOLS), 4, 2) AS CS
,SUBSTR(P.ACCESSNAME, 1, 12) AS INDEX
,P.INDEXONLY AS IO
,SUBSTR(CASE SORTN_UNIQ WHEN 'N' THEN '-' ELSE 'Y' END ||
CASE SORTN_JOIN WHEN 'N' THEN '-' ELSE 'Y' END ||
CASE SORTN_ORDERBY WHEN 'N' THEN '-' ELSE 'Y' END ||
CASE SORTN_GROUPBY WHEN 'N' THEN '-' ELSE 'Y' END
, 1 , 4) AS UJOG
,SUBSTR(CASE SORTC_UNIQ WHEN 'N' THEN '-' ELSE 'Y' END ||
CASE SORTC_JOIN WHEN 'N' THEN '-' ELSE 'Y' END ||
CASE SORTC_ORDERBY WHEN 'N' THEN '-' ELSE 'Y' END ||
CASE SORTC_GROUPBY WHEN 'N' THEN '-' ELSE 'Y' END
, 1 , 4) AS UJOGC
,P.PREFETCH AS P
,P.COLUMN_FN_EVAL AS CE
,P.PAGE_RANGE AS PR
,P.QBLOCK_TYPE AS TYPE
,P.MERGE_JOIN_COLS AS MJC
,S.PROCMS AS MS
,S.PROCSU AS SU
,D.EARLY_OUT AS EO
,D.BLOCK_FETCH AS BF
,F.ORDERNO AS ON
,F.PREDNO AS PN
,F.STAGE AS STAGE
,F.PUSHDOWN AS PD
,R.TYPE AS TYPE
,R.ADDED_PRED AS AP
,R.CLAUSE AS CLAUSE
,R.ORIGIN AS OR
,R.REDUNDANT_PRED AS RP
,R.TEXT AS TRANSFORMED_PREDICATE
FROM PLAN_TABLE P
FULL OUTER JOIN
DSN_STATEMNT_TABLE S
ON P.QUERYNO = S.QUERYNO
AND P.APPLNAME = S.APPLNAME
AND P.PROGNAME = S.PROGNAME
AND P.COLLID = S.COLLID
AND P.GROUP_MEMBER = S.GROUP_MEMBER
AND P.SECTNOI = S.SECTNOI
AND P.VERSION = S.VERSION
AND P.EXPLAIN_TIME = S.EXPLAIN_TIME
FULL OUTER JOIN
DSN_DETCOST_TABLE D
ON P.QUERYNO = D.QUERYNO
AND P.APPLNAME = D.APPLNAME
AND P.PROGNAME = D.PROGNAME
AND P.COLLID = D.COLLID
AND P.GROUP_MEMBER = D.GROUP_MEMBER
AND P.SECTNOI = D.SECTNOI
AND P.VERSION = D.VERSION
AND P.EXPLAIN_TIME = D.EXPLAIN_TIME
AND P.QBLOCKNO = D.QBLOCKNO
AND P.PLANNO = D.PLANNO
FULL OUTER JOIN
DSN_FILTER_TABLE F
ON P.QUERYNO = F.QUERYNO
AND P.APPLNAME = F.APPLNAME
AND P.PROGNAME = F.PROGNAME
AND P.COLLID = F.COLLID
AND P.GROUP_MEMBER = F.GROUP_MEMBER
AND P.SECTNOI = F.SECTNOI
AND P.VERSION = F.VERSION
AND P.EXPLAIN_TIME = F.EXPLAIN_TIME
AND P.QBLOCKNO = F.QBLOCKNO
AND P.PLANNO = F.PLANNO
FULL OUTER JOIN
DSN_PREDICAT_TABLE R
ON F.QUERYNO = R.QUERYNO
AND F.APPLNAME = R.APPLNAME
AND F.PROGNAME = R.PROGNAME
AND F.COLLID = R.COLLID
AND F.GROUP_MEMBER = R.GROUP_MEMBER
AND F.SECTNOI = R.SECTNOI
AND F.VERSION = R.VERSION
AND F.EXPLAIN_TIME = R.EXPLAIN_TIME
AND F.QBLOCKNO = R.QBLOCKNO
AND F.PREDNO = R.PREDNO
WHERE 1 = 1
AND P.QUERYNO IN (1 , 2 )
ORDER BY 1 , 2 , 3 , 4 , 5 , 24 , 25
;

Here I limit it to just the QUERYNO 1 and 2 as these were the numbers used for the EXPLAIN command:

EXPLAIN ALL SET QUERYNO = 1 FOR              
SELECT INSTANCE, CLONE
FROM SYSIBM.SYSTABLESPACE A
WHERE ( SELECT B.TSNAME
FROM SYSIBM.SYSTABLES B
,SYSIBM.SYSINDEXES C
WHERE C.CREATOR = ?
AND C.NAME = ?
AND C.TBCREATOR = B.CREATOR
AND C.TBNAME = B.NAME ) = A.NAME
AND A.DBNAME = ?
;
EXPLAIN ALL SET QUERYNO = 2 FOR
SELECT A.INSTANCE, A.CLONE
FROM SYSIBM.SYSTABLESPACE A
,SYSIBM.SYSTABLES B
,SYSIBM.SYSINDEXES C
WHERE C.CREATOR = ?
AND C.NAME = ?
AND C.TBCREATOR = B.CREATOR
AND C.TBNAME = B.NAME
AND A.DBNAME = ?
AND A.NAME = B.TSNAME
AND A.DBNAME = B.DBNAME
WITH UR
;
--------+--------+------+--------+---------+---------+--------+------+---
PROGNAME LINE QNO PNO SQ M TABLE_NAME A PA CS INDEX IO UJOG UJOGC
--------+--------+------+--------+---------+---------+--------+------+---
DSNESM68 00001 01 01 00 0 SYSTABLESPACE I 02 DSNDSX01 N ---- ----
DSNESM68 00001 01 01 00 0 SYSTABLESPACE I 02 DSNDSX01 N ---- ----
DSNESM68 00001 02 01 00 0 SYSINDEXES I 02 DSNDXX01 N ---- ----
DSNESM68 00001 02 01 00 0 SYSINDEXES I 02 DSNDXX01 N ---- ----
DSNESM68 00001 02 02 00 1 SYSTABLES I 02 DSNDTX01 N ---- ----
DSNESM68 00001 02 02 00 1 SYSTABLES I 02 DSNDTX01 N ---- ----
DSNESM68 00002 01 01 00 0 SYSINDEXES I 02 DSNDXX01 N ---- ----
DSNESM68 00002 01 01 00 0 SYSINDEXES I 02 DSNDXX01 N ---- ----
DSNESM68 00002 01 02 00 1 SYSTABLES I 02 DSNDTX01 N ---- ----
DSNESM68 00002 01 02 00 1 SYSTABLES I 02 DSNDTX01 N ---- ----
DSNESM68 00002 01 02 00 1 SYSTABLES I 02 DSNDTX01 N ---- ----
DSNESM68 00002 01 03 00 1 SYSTABLESPACE I 02 DSNDSX01 N ---- ----
DSNESM68 00002 01 03 00 1 SYSTABLESPACE I 02 DSNDSX01 N ---- ----
DSNE610I NUMBER OF ROWS DISPLAYED IS 13
--+---------+---------+---------+---------+---------+---------+--------
P CE PR TYPE MJC MS SU EO BF ON
--+---------+---------+--------+---------+---------+---------+---------
SELECT ------ 1 2 N N 1
SELECT ------ 1 2 N N 2
NCOSUB ------ 1 2 N N 1
NCOSUB ------ 1 2 N N 2
NCOSUB ------ 1 2 N N 1
NCOSUB ------ 1 2 N N 2
SELECT ------ 1 2 N N 1
SELECT ------ 1 2 N N 2
SELECT ------ 1 2 N N 1
SELECT ------ 1 2 N N 2
SELECT ------ 1 2 N N 3
SELECT ------ 1 2 N N 1
SELECT ------ 1 2 N N 2
---------+---------+---------+---------+---------+---------+---------+-----
PN STAGE PD TYPE AP CLAUSE OR RP TRANSFORMED_PREDICATE
---------+---------+---------+---------+---------+---------+---------+-----
3 MATCHING EQUAL WHERE U N "A"."DBNAME"=(EXPR)
2 MATCHING EQUAL WHERE U N "A"."NAME"=(SELECT "B"."TSNAME"
5 MATCHING EQUAL WHERE U N "C"."CREATOR"=(EXPR)
6 MATCHING EQUAL WHERE U N "C"."NAME"=(EXPR)
7 MATCHING EQUAL WHERE U N "C"."TBCREATOR"="B"."CREATOR"
8 MATCHING EQUAL WHERE U N "C"."TBNAME"="B"."NAME"
2 MATCHING EQUAL WHERE U N "C"."CREATOR"=(EXPR)
3 MATCH EQUAL WHERE U N "C"."NAME"=(EXPR)
4 MATCH EQUAL WHERE U N "C"."TBCREATOR"="B"."CREATOR"
5 MATCHING EQUAL WHERE U N "C"."TBNAME"="B"."NAME"
9 STAGE1 EQUAL T WHERE N "B"."DBNAME"=(EXPR)
6 MATCHING EQUAL WHERE U N "A"."DBNAME"=(EXPR)
7 MATCHING EQUAL WHERE U N "A"."NAME"="B"."TSNAME"

It shows you a lot more data than the normal bare bones Explain PLAN_TABLE data especially here the TYPE, STAGE and ADDED_PREDICATE (AP Column).

Feel free to send me your comments and ask questions.

TTFN,
Roy Boxwell
Senior Architect

2019-02 Global Temporary Confusion

Global Temporary Table usage in Db2 z/OS

Learn, through examples, why CGTT (CREATEd Global Temporary Tables) are possibly better than DGTT (DECLAREd Global Temporary Tables) and how to use them to speed up performance.

When Global Temporary Tables were first introduced (DB2 V5.1) it was all pretty clear! You used normal DDL to CREATE a GLOBAL TEMPORARY Table that existed in the Db2 Catalog, but every process got “their” own copy. There were limitations like No UPDATE, No Single DELETE, no Default Values and ROLLBACK / COMMIT and data reuse was not simple but it all sort of worked…

Global – but not as you know it!

Then along came the brand new Global Temporary Table (DB2 V6.1) which, just to make matters clearer, they prefixed with DECLARE so that it was 100% obvious that it had *nothing* to do with the “other” temporary table… I tell ya, if I ever meet the developer who came up with these names in a dark alley one night…. But I digress… So with the new DECLARE version you suddenly *could* do Update and Delete and it came with an ON COMMIT clause to make it simpler to handle data with COMMIT – Hoorah!

Global Temporary Table in Db2 12 – All new and improved

As of Db2 12, the full list of things you can do with a CREATE GLOBAL TEMPORARY TABLE (from now on simply CGTT) as opposed to a DECLARE GLOBAL TEMPORARY TABLE (from now on simply DGTT) is:

 What            CGTT           DGTT
MERGE No Yes
UPDATE No Yes
Single DELETE No Yes
DEFAULTs No Yes
ROLLBACK Deletes all Yes (Plus NOT LOGGED ON ROLLBACK)
COMMIT Deletes all Depends on ON COMMIT
(unless WITH HOLD)
Indexes? No Yes
WITH RETURN Yes Yes <- This is very good news
TO CLIENT

CGTT why?

So, looking at the list, you have to wonder – “Why would anyone be using CGTTs?”
Well the reason is “Performance!” –
DGTTs might do all the tricks but they also run like a dog!

Belief is ok, test is trust!

I wrote a couple of small COBOL programs that did the same thing: one using a DGTT and one using a CGTT. All the programs did, was call a section 1000 times that DECLARED the table (or obviously not!) and then called another section that inserted 10 rows and opened a cursor to then fetch these rows right back with an ORDER BY. I chose this test as most DGTT/CGTT usage is low-volume but high called rate (Think stored procedures here!) so I wanted to see what sort of overhead 1000 DGTTs caused.

No EDMPOOL here!

Monitoring is not easy as DGTT usage is not in the EDMPOOL, so I did it the “good old way” by running my tests when I was alone on the system and five times each. I was a bit shocked by the results…

What               CGTT                   DGTT
DBM1 EXCPs      0               8,817
MSTR EXCPs      0              59,653
Db2 CPU         0.140 secs       3.546 secs
Program cpu      4.506 secs     101.124 secs
Program elapsed 4.958 secs     118.730 secs

I then changed the DGTT to also be NOT LOGGED

What                   DGTT 
DBM1 EXCPs           66
MSTR EXCPs      47
Db2 CPU           2.446 secs
Program cpu     100.818 secs
Program elapsed 116.426 secs

Not that much better! But remember, I did not do any DELETE or UPDATE processing so your performance data may be better than mine.

Impressive!

Now the reason for all of the I/O and CPU is
all of the internals that Db2 must do for the DGTT, remember the CGTT has *already* been declared and exists in the catalog. The DGTT naturally not…
This can, as be seen, add a lot of overhead!

ROT still correct

Now the good old Rule of Thumb still holds true:

A CGTT is great for doing zero updates and sequential access!

Db2 10 improvement

When a cursor is declared with the attribute RETURN TO CLIENT then this cursor is available from the originating calling client even if the current stored procedure is down line. This is very neat as the cursor is “invisible” to all the intermediate procedures.

Death by DGTT

The problem in the past was the “hand shaking” of the final stored procedure result set. It had to be read out and put into another DGTT and then this handling was repeated all the way back up the calling chain. Now, if you have a “final cursor”, you can simply declare it as RETURN TO CLIENT – This saves tons of CPU and elapsed time!

Final tip!

Remember if you use STATIC SCROLLABLE cursors then they *must* use DGTTs in the background to actually work!

Even more work…

Then a colleague who proof-reads all my newsletters, a nasty job but someone must do it, asked me:

“I wonder what the numbers would be if you declared the GTT only once, then INSERTED/SELECTED/DELETED a bunch of rows 1000 times? Because that’s how I’m using it …“

So then I changed the programs to only DECLARE once, insert a 1000 rows, Select a 1000 rows and then do a MASS DELETE of the data.

What             CGTT           DGTT            DGTT NOT LOGGED
DBM1 EXCPs      7,572         7,720           7,769
MSTR EXCPs      0               672           238
Db2 CPU         0.220 secs     0.310 secs      0.410 secs
Program cpu     0.346 secs     0.482 secs      0.468 secs
Program elapsed 0.704 secs     0.836 secs      0.818 secs

Much closer, in terms of performance, but still the CGTT is quicker – Time to check your usage of these Temporary Tables if you ask me!


Feel free to send me your comments and ask questions.

TTFN,
Roy Boxwell
Senior Architect