Themis Top 25 Db2 Tuning Presentation

7/25/2019 Themis Top 25 Db2 Tuning Presentation

1/58

Top 25+ DB2 SQLTuning Tips for

Developers

Presented by Tony Andrews, Theis !n"#t$ndrews%theisin"#"o
mailto:[email protected]:[email protected]


2/58

&b'e"tives

By the end of this present$tion, developers should be $ble to(

) *nderst$nd wh$t SQL &ptii$tion is

) *nderst$nd St$ge versus St$ge 2 predi"$tes) *nderst$nd wh$t the DB2 -.pl$in Tool does

) *nderst$nd wh$t the /isu$l -.pl$in "$n tell us

) *nderst$nd wh$t $0es 1ueries run too long) Le$rn wh$t to do when they see $ t$ble s"$n t$0ing pl$"e#

) Le$rn steps to tuning $ 1uery


3/58

3y -.perien"e Shows

) Th$t the $'ority of perfor$n"e probles $ong$ppli"$tions $re "$used by poorly "oded progr$s oriproperly "oded SQL#

) Th$t poor perforing SQL is responsible for $s u"h $s4 per"ent of responsetie issues#

) Th$t ost developers need edu"$ted in how to re$d $nd$n$lye the DB2 -.pl$in output#

) There $re $ny SQL developers in the !T industry, but !h$ve found th$t less th$n 6 of the re$lly 0now the

perfor$n"e issues involved with SQL progr$ing orhow to fi. the#


4/58


5/58

Developers should $lw$ys progr$ with

perfor$n"e in ind#

Progr$ers should $lw$ys h$ve two go$ls in ind when developing

progr$s $nd $ppli"$tions for user groups#

) To get the "orre"t results for re1uested d$t$

) To get the results b$"0 $s 1ui"0ly $s possible

So $ny ties progr$ers tod$y lose sight of the se"ond go$l# They

either(

) Do not 0now wh$t to do to get progr$s to run f$ster

) Bl$e other pie"es of their environent 7d$t$b$se, networ0, T8P9!P,wor0st$tion, oper$ting syste, et"#:, or

) Thin0 th$t the tie spent pro"essing the d$t$ w$s pretty good b$sed on

the $ount of d$t$#


6/58

The DB2 &ptiier

Catalog StatisticsObject Definitions

Access Path

Access Path Hint

RID Pool Sorts


7/58

St$ge versus St$ge 2 Predi"$tes

) St$ge 7DB2 D$t$ 3$n$ger: is responsible for tr$nsl$ting the d$t$

stored on p$ges into $ result set of rows $nd "oluns# Predi"$tes

th$t $re written in $ f$irly str$ightforw$rd w$y "$n usu$lly be

ev$lu$ted by the D$t$ 3$n$ger with rel$tively little e.pense#

) St$ge 2 7;el$tion$l D$t$ Servi"es: h$ndle ore "ople. predi"$tes,

d$t$ tr$nsfor$tions, $nd "oput$tions# These St$ge 2 predi"$tes

$re u"h ore e.pensive for DB2 to resolve th$n St$ge due to

$ddition$l pro"essing $nd $ddition$l "ode p$th# Addition$lly, ;DS

"$nnot $0e effe"tive use of inde.es#


8/58

Top 25+ Tuning Tips


9/58

Top 2+ Tuning Tips ;&3 -3PL&@-- -;- @-A;7!;-DAT-: C 25

Should be "oded $s(

S-L-8T -3P?&, LAST?A3- >;&3 -3PL&@--

-;- !;-DAT- B-T--? 25E $nd 252FE

V&' Can no( create indexes on S)* expressions#


10/58

Top 25+ Tuning Tips ;&3 -3PL&@-- >;&3 -3PL&@--

-;- SALA;@ H 5# 9 # -;- !;-DAT- H DAT-7K:

+ J DA@S

V&' Can no( create indexes on S)* expressions


11/58

Top 25+ Tuning Tips


12/58

Top 25+ Tuning Tips irst 9 &rder By within sub1ueries#


13/58

Top 25+ Tuning Tips


14/58

Top 25+ Tuning Tips


15/58

Top 25+ Tuning Tips


16/58

Top 25+ Tuning Tips


17/58

Top 25+ Tuning Tips


18/58

Top 25+ Tuning Tips


19/58

Top 25+ Tuning Tips


20/58

Top 25+ Tuning Tips ". Only code the columns needed in the Select portion o the S)* statement.

$ving e.tr$ "oluns "$n h$ve $n $ffe"t on(

The optiier "hoosing !nde. &nlyE -.pensiveness of $ny sorts

&ptiierEs "hoi"e of 'oin ethods


21/58

Top 25+ Tuning Tips


22/58

Top 25+ Tuning Tips


23/58

Top 25+ Tuning Tips or -.$ple(

Non Correlated Su3uery Can also e coded as'

S-L-8T -#-3P?&, -#LAST?A3- S-L-8T -#-3P?&, -#LAST?A3-

>;&3 -3PL&@-- - >;&3 -3PL&@-- - -;- -#-3P?& !? -;- -U!STS

7S-L-8T D#3;?& 7S-L-8T

>;&3 D-PA;T3-?T D >;&3 D-PA;T3-?T D

-;- D#D-PT?& L!- D6V: -;- D#3;?& C -#-3P?&

A?D D#D-PT?& L!- D6E:

V&' 8loal )uery Optimization. Optimizer no( tries to determine ho( an access path o one

3uery loc$ may aect the others. #his can e seen at times y D12 re(riting an ,Exists-

su3uery into a @oin0 or an ,?n- su3uery into an ,Exists- su3uery .

V& optimization calls this ,Correlating- and ,DeAcorrelating-.


24/58

Top 25+ Tuning Tips


25/58

Top 25+ Tuning Tips


26/58

Top 25+ Tuning Tips


27/58

Top 25+ Tuning Tips


28/58

Top 25+ Tuning Tips


29/58

Top 25+ Tuning Tips


30/58

Top 25+ Tuning Tips re1uen"y /$lue st$tisti"s on $ny "olun7s: with uneven distribution roup st$tisti"s 7distin"t v$lues over $ group of "oluns: for $ny set of "oluns th$t

$re "orrel$ted

V&' Qu$ntile St$tisti"s 7further bre$0down of st$tisti"s:# elps with r$nge predi"$tes,

between predi"$tes, $nd the li0e predi"$te# -spe"i$lly where there e.ists ot SpotsEof d$t$ distribution#


31/58

Top 25+ Tuning Tips


32/58

Top 25+ Tuning Tips or e.$ple( There e.ists $ t$ble with illion rows# !n this t$ble e.ists $n inde. on

the "olun St$tus8ode# After $ typi"$l ;unst$ts utility is e.e"uted $g$inst this t$ble,

the optiier will 0now th$t there $re F different v$lues for the st$tus "ode# After $

spe"i$l ;unst$ts th$t spe"ifies fre1uen"y v$lue st$tisti"s for th$t "olun, DB2 will 0now

the following d$t$ distributions(

St$tus 8ode v$lue AE "ont$ins 56 of the d$t$

St$tus 8ode v$lue BE "ont$ins M56 of the d$t$

St$tus 8ode v$lue 8E "ont$ins 56 of the d$t$#

A progr$ "ont$ins the following SQL# The v$lue is $lw$ys 8E th$t "oes into

the progr$(

S-L-8T 8&L, 8&L2, 8&LF

>;&3 TABL-

-;- STAT*S8D C (&ST/A;!ABL-


33/58

Top 25+ Tuning Tips ;&3 -3PL&@-- -

-;- -U!STS

7S-L-8T

>;&3 D-PA;T3-?T D

-;- D#3;?& C -#-3P?&

A?D D#D-PT?& L!- D6E:


34/58

Top 25+ Tuning Tips


35/58

Top 25+ Tuning Tips


36/58

Top 25+ Tuning Tips or e.$ple(

S-L-8T -3P?&, LAST?A3-, D-PT?&, SALA;@

>;&3 -3PL&@--

-;- D-PT?& H K

&PT!3!W- >&; M ;&S

!f you 0now for sure th$t your s"reen will only pro"ess M rows no $tter how

$ny get returned, then the SQL st$teent should spe"ify >et"h >irst M ;ows

&nlyE#


37/58

Top 25+ Tuning Tips . ?nsert (ith Select V6".

DB2 /4 introdu"ed the $bility to now sele"t wh$t w$s 'ust inserted with the s$est$teent s$ving ultiple "$lls to DB2# This $g$in we "$ll ;el$tion$lE progr$ing

inste$d of Pro"edur$lE progr$ing# The st$teent "$n retrieve the following

infor$tion#

!dentity "oluns or se1uen"e v$lues th$t get $uto$ti"$lly $ssigned by DB2

*serdefined def$ults $nd e.pressions th$t $re not 0nown to the developer 8oluns odified by triggers th$t "$n v$ry fro insert to insert depending on v$lues

;&!Ds, 8*;;-?T T!3-STA3P th$t $re $ssigned $uto$ti"$lly

>or e.$ple(

S-L-8T 8, 82, 85, 8O, 84, 8, 82 >;&3 >!?AL TABL-

7!?S-;T 782, 8F, 8M, 85, 8O, 84, 8:

/AL*-S 7AB8E, 2F#2, AE, D->E,

5#, !E, 24E:

:


38/58

Top 25+ Tuning Tips ;&3 -3PL&@-- -

-;- ?&T -U!STS

7S-L-8T

>;&3 D-PA;T3-?T D

-;- D#3;?& C -#-3P?&:


39/58

Top 25+ Tuning Tips !;ST?A3-, LAST?A3-,

D-PA;T3-?T:

&? -#-3P?& C ?-!T-3#-3P?&

-? 3AT8-D T-?

*PDAT- S-T >!;ST?A3- C ?-!T-3#>!;ST?A3-,

LAST?A3- C ?-!T-3#LAST?A3-

-? ?&T 3AT8-D T-?

!?S-;T 7-3P?&, >!;ST?A3-, LAST?A3-, D-PA;T3-?T:

/AL*-S 7?-!T-3#-3P?&, ?-!T-3#>!;ST?A3-,

?-!T-3#LAST?A3-, ?-!T-3#D-PA;T3-?T:


40/58

Top 25+ Tuning Tips et"h the ;&!D 7;!D $s of /N: for e$"h row being pro"essed in the

;e$d &nlyE "ursor, $nd e.e"ute $ll upd$te or delete st$teents using the;&!D9;!D v$lue in pl$"e of the 0ey fields for better perfor$n"e#


41/58

Top 25+ Tuning Tips


42/58

Top 25+ Tuning Tips


43/58

Top 25+ Tuning Tips $"ility

F: -.tern$l SPs needs $ l$ngu$ge environent in whi"h to e.e"ute# This is

provided by $ L3$n$ged $ddress sp$"e# Thre$ds get suspended $s

T8Bs get $ssigned#

M: hen $ n$tive SP is "$lled, DB2 swit"hes fro the "$llers p$"0$ge to the

stored pro"edure p$"0$ge# ?o thre$d suspension $nd t$s0 swit"hing, $nd

no del$y in stored pro"edure e.e"ution


44/58

Top 25+ Tuning Tips un"tion 3ode

J: SQL "ode runs $s $ interpretive l$ngu$ge $nd is in"luded $s p$rt of the

p$"0$ge# $t"h out for SPs th$t "ont$in $ lot of "ode#

4: -.tern$l 7to DB2: resour"es $re not re1uired to prep$re or to e.e"ute $

n$tive SQL pro"edure# >or e.$ple "opiles $nd lo$d libr$ries#

N: ?o L3 $ppli"$tion e.e"ution environent( when $ n$tive SQL pro"edure

is "$lled it e.e"utes in the DB2 d$t$b$se servi"es $ddress sp$"e 7DB3:#


45/58

?$tive SP -.$ple


46/58

?$tive SP -.$ple


47/58

?$tive SP -.$ple A*LT X

D-8LA;- " 8*;S&; >&;

S-L-8T s$l$ry >;&3 st$ff &;D-; B@ s$l$ryX

D-8LA;- "2 8*;S&; !T ;-T*;? >&;

S-L-8T n$e, 'ob, s$l$ry

>;&3 st$ff-;- s$l$ry H edi$nS$l$ry

&;D-; B@ s$l$ryX

D-8LA;- -U!T A?DL-; >&; ?&T >&*?D

S-T edi$nS$l$ry C X

S-L-8T 8&*?T7G: !?T& vnu;e"ords >;&3 STA>>X

&P-? "X

!L- v"ounter R 7vnu;e"ords 9 2 + : D&>-T8 " !?T& edi$nS$l$ryX

S-T v"ounter C v"ounter + X

-?D !L-X

8L&S- "X

&P-? "2X

-?D


48/58

Ten Steps to Tuning $ Query

: 8he"0 every predi"$te# Are they inde.$ble $nd St$ge K2: !s there $ Distin"tEK !s it neededK 8$n it be rewrittenK Try roup B@E#

F: Are there sub1ueriesK ;ewrite !nE $s -.istsE $nd vi"e vers$#

M: 8he"0 DB2 st$tisti"s on every t$ble $nd every "olun involved#

5: 8he"0 the nuber of ties every SQL st$teent is getting e.e"uted# 8$n the

logi" be "h$nged to "ut down the nuber of ties re1uests $re sentKO: 8he"0 the DB2 -.pl$in# Are there $ny t$ble s"$nsK Any sortsK

J: 8he"0 the DB2 /-# Are there $ny predi"$tes with "r$y >ilter >$"torsK

4: 8he"0 the DB2 -.pl$in# !f there is $ 'oin, wh$t is the order of t$bles "hosenK

N: Are there $ny "orrel$ted sub1ueriesK 8$n they be !nde.&nlyK

: Are there $ny "oluns in predi"$tes with uneven distribution st$tisti"sK Shouldthe v$lue be h$rd "odedK

: Are there $ny r$nge predi"$tes# 8ould histogr$ st$tisti"s helpK


49/58

?ew DB2 /4 SQL >e$tures

9ollo(ing are some o the ne( S)* eatures in D12 V6'

: 3ore St$ge predi"$tes

2: 3ulti ;ow >et"h, *pd$te, $nd !nsert

F: 3ultiple Distin"ts

M: -.pressions in the roup ByE

5: 8oon T$ble -.pression

O: Dyn$i" S"roll$ble 8ursors

J: Se1uen"es versus !dentity 8oluns

4: 3$teri$lied Query T$bles 73QTs:

N: ;e"ursive SQL

: 3ore effi"ient use of inde.es# >orw$rd $nd B$"0w$rd s"$ns: ?ew U3L fun"tions $nd d$t$types

2: ?ew et Di$gnosti"sE for w$rning $nd error infor$tion

F: Sele"t fro $n !nsert st$teent

M: S"$l$r >ullsele"t within $ Sele"tE, 8$seE, *pd$te, et"#


50/58

?ew DB2 /N SQL >e$tures

9ollo(ing are some o the ne( S)* eatures in D12 V&'

: Set oper$tions !nterse"tE $nd -."eptE

2: 3erge st$teent for *psertE pro"essing# !nsert or *pd$te

F: &LAP fe$tures for ;$n0ing $nd ?ubering of d$t$

M: ?$tive SQL Stored Pro"edures

5: !nste$d ofE Triggers

O: ?ew support $nd SQL for U3L d$t$

J: &ptii$tion Servi"e 8enter 7&S8:

4: Distin"t sort $void$n"e with non uni1ue inde.es

N: !nde.ing on -.pressions

: St$tisti"s on /iews: S0ipped lo"0ed d$t$

2: Trun"$te st$teent


51/58

?ew DB2 /N SQL >e$tures

9ollo(ing are some o the ne( S)* eatures in D12 V&'

F: Arr$y host v$ri$bles now supported for stored pro"edures

M: Tiest$p $uto upd$te for inserts $nd *pd$tes

5: &ptiisti" lo"0ing

O: ?ew D-8>L&AT d$t$type

J: Sele"t fro *pd$te or Delete getting old or new v$lues

4: >et"h >irst, &rder B@ within sub1ueries

N: ;-&PT A*T& 7Dyn$i" SQL:

2: D$t$ Studio for ?$tive Stored Pro"edures


52/58

-.perien"e Shows

Th$t $ little 0nowledge goes $ long w$y in !T developent shops

When you know something, say what you know. When you don't know something, say that you

don't know. That is knowedge.!


53/58

-.perien"e Shows

Th$t good tr$ining goes $ long w$y in !t developent shops#

The mediocre teacher tes.

The good teacher e"pains.The superior teacher demonstrates.

The great teacher inspires.!

i h


54/58

-.perien"e Shows

Th$t h$ndson "onsulting $longside developers goes $ long w$yin !T developent shops

#onsuting engagements ne$er seem compete

uness there is a comfortabeness that the cientstaff fuy understands a issues, and has theabiity to mo$e forward on their own.!

i Sh


55/58

-.perien"e Shows

Th$t st$nd$rds, guidelines, $nd w$l0throughs s$ves 8P* "osts

$nd in"ident reporting in !T shops#

There is aways time for a %& minute wakthrough!

T i i d 8 lti 8h 0 & t


56/58

Tr$ining $nd 8onsulting# 8he"0 &ut

www#theisin"#"o

) &nsite $nd Publi"

) !nstru"tor led

)$ndson) 8ustoi$tion

) -.perien"e

) &ver 25 DB2 "ourses

) &ver M !T "oursesUS !A6>>A;7:A+>>>

?ntl. !A&>6A2++A6&>>

-du"$tion 8he"0 out


57/58

9inallyG oo$ o D12 S)* tuning tips or

developers0 speciically designed to improveperormance.

D12 S)* developers no( have a handy

reerence guide (ith tuning tips to improve

perormance in 3ueries0 programs and

applications.

-du"$tion# 8he"0 out

www#db2s1ltuningtips#"o


58/58

Th$n0 @ou for $llowing us $t Theis to

sh$re soe of our e.perien"e $nd

0nowledgeY

) e hoped th$t you le$rned soething new tod$y) e hope th$t you $re $ little ore inspired to "ode with

perfor$n"e in ind

ha$e noticed that when the de$eopers get

educated, good ()* programming standards are in

pace, and program wakthroughs are e"ecutedcorrecty, incident reporting stays ow, #+ costs do not get

out of contro, and most performance issues are foundbefore promoting code to production.!

Themis Top 25 Db2 Tuning Presentation

Documents