Americas Headquarters
EMEA Headquarters
Asia-Pacific Headquarters
100 California Street, 12th Floor San Francisco, California 94111
York House 18 York Road Maidenhead, Berkshire SL6 1SF, United Kingdom
L7. 313 La Trobe Street Melbourne VIC 3000 Australia
The High Performance SQL Server DBA
Managing Multiple Database Platforms for Performance & Availability
By Scott Walz, Sr. Director of Product Management at Embarcadero Technologies and Stephen Wynkoop, Microsoft SQL Server MVP and Founder of SSWUG.org
July 2010
Embarcadero
TABLE Introduct Understa
VisualizData N
Optimizin
CatchinIdentifySQL Tu
Storage M Performa
BottlenWorkloRatio A
Capacity Change M Conclusio About Em About th
o Technologie
E OF CO
tion .........
anding thezing the Dormalizat
ng Poor-Png the Proying the Muning Tech
Managem
ance Mananeck/Respoad AnalysAnalysis ...
Managem
Managem
on ...........
mbarcade
e Authors
es
ONTEN
...............
e DatabasDatabase .tion vs. Pe
Performinoblem EarMain Sourhniques ..
ment ........
agement ponse Timsis ..........................
ment .......
ment .........
...............
ero and SS
s .............
NTS
................
se Design................erformanc
g SQL .....rly ...........rce of the................
................
...............me Analysi................................
................
................
................
SWUG.org
................
...............
n .............................ce ...........
...............
............... Bottlene...............
...............
...............is ..........................................
...............
...............
...............
g ............
...............
The H
...............
...............
...............
...............
...............
...............eck ..........................
...............
...............
...............
...............
...............
...............
...............
...............
...............
...............
High Perform
................
................
................
................
................
................
................
................
................
................
................
................
................
................
................
................
................
................
ance SQL Se
...............
...............
...............
...............
...............
...............
...............
...............
...............
...............
...............
...............
...............
...............
...............
...............
...............
...............
rver DBA
- 1 -
... - 2 -
... - 3 -
... - 3 -
... - 4 -
... - 5 -
... - 5 -
... - 5 -
... - 6 -
... - 8 -
. - 10 -
. - 10 -
. - 11 -
. - 11 -
. - 12 -
. - 14 -
. - 14 -
. - 15 -
. - 15 -
Embarcadero
INTRO SQL Server only mounte
• MulteachNewdom
• Morpetamadcom
• Mordatanummanexte
• Conrise tplanmore
• AddDBAdatadynaadvaimprhuminformultincre
• Not freedtheirwhicto be
Regardless database symany sourceshould inclu
• U
o Technologie
ODUCTI
DBAs have ed in recent
tiple SQL Seh release, anw features, sumain tradition
e Data – Daabytes of datde common d
pliance mor
e Databaseabases, most
mber, so haveage complend one’s SQ
solidation /to new kindsning, to pere.
ded CompleA’s daily workabases whichamic access ancements srovements, D
man interventrmation withtiple databasease the com
Enough Pedom to hire,r critical datach in turn pute more prod
of these chaystems availaes, and you
ude the follo
Understandi
es
ION always tackleyears. Seve
erver Versiod SQL Serve
uch as CLR fnally reserve
ata growth ista. This growdatabase mare complex.
es – While sot manage doe database pex heterogenQL Server ski
/ Virtualizatis of issues informance m
exity – Whilek, Microsoft h are increasto system heuch as MicroDBMS envirotion and manh each new dse versions,
mplexity of t
eople – Expe, many organabases. This ts the onus o
ductive and
allenges, as aable and optneed a strat
owing compo
ng the Data
ed complex ral trends im
ons – Microser DBAs are unctions, ared for develo
s exploding. wth has creatanagement
ome DBAs uozens and evplatform typeneous databills into othe
ion – The run database a
monitoring an
e all of the aband other d
singly “self-mealth informosoft’s databonments arenagement. D
database releand adding he analysis n
erienced DBnizations cantranslates in
on organizat“scalable.”
an overworktimized for htegy that wilonents:
base Design
tasks amidsmpacting SQ
soft SQL Serrequired to
e stretching opers.
DBAs increted a relentltasks such a
se SQL Servven hundredes and versio
base environer database t
sh to consoadministrationd troublesh
bove trends database venmanaged” oation and aubase tuning e growing mDBAs have tease. This camultiple da
needed.
As have alwnnot find thento a highertions to find
ked SQL Servhigh performl contribute
n
The H
st shifting prQL Server DB
ver continuebe experts DBAs from
asingly maness demands back-ups,
ver only, or mds. As databons. Today, ments, makitypes.
lidate and von, from virtuhooting, to in
are adding ndors have br “self-tuninutomated anand suggesore complexo quickly aban be espectabase platf
ways been hae talent or exratio of dataways that w
ver DBA, yomance. Perfo
the most to
High Perform
iorities, and As:
es to evolve on *all* verstheir comfo
age terabyted for increasecapacity pla
manage onlyases grow inmany SQL S
ing it more c
irtualize dataualized servencorporating
tremendousbeen respong” through
nalysis tools.tion tools arx and requir
bsorb and macially acute wforms can ex
ard to find. Expertise theyabases mana
will empower
u are expectrmance prohigh availab
ance SQL Se
pressures h
and advancsions of SQLrt zone into
es and evened storage a
anning, and
y a handful on both size aServer DBAschallenging
a centers haer capacity g SQL Azure
s complexityding with the use of m. While re time-savine knowledgaster new
when workingxponentially
Even with incy need to maged per DBr SQL Server
ted to keep blems comebility. This st
rver DBA
- 2 -
ave
e with L Server. a
and has SLA
of and s to
s given
e, and
y to a
more
ng eable
g with
creasing anage
BA, r DBAs
key e from trategy
Embarcadero
• C
• M
• M
• M
• M
This paper wto deliver hi
UNDEAt some poWhether thean app deve A great waymodel. Mosto visualize t“seeing” threports for oelements.
VISUALIZA logical daacross multiallows you tterms and ruimplementa
• R
• O
• P
• R
Whether yowith a thoromodelers dumodel and i
o Technologie
Catching Po
Managing St
Managing Pe
Managing C
Managing C
will outline tigh database
ERSTANint, all DBAse new databeloper, supp
y to become st data modethe table stre database other users,
ZING THata model is iple databasto use busineules enforceations. Logic
Reducing da
Organizing b
Providing a f
Reducing de
u inherit an ough understuring the deits potential
es
or-Performin
torage
erformance
apacity
hange
techniques te performan
NDING s will be requ
base comes vporting a new
more intimaeling tools aructure, relaton paper, yoand analyze
E DATABan invaluabl
se platformsess names fo
ed in the mocal modeling
ata redundan
business req
foundation o
evelopment
existing dattanding of th
evelopment pimpact on p
ng SQL Early
o streamlinence and avail
THE Duired to supvia a new apw database
ate with a nellow DBAs ttionships, anou can unde the consiste
BASE le tool for m. The logicalor tables anddel while mag helps DBA
ncy
quirements
of the datab
and mainten
abase or buhe databasephase, you wperformance
y
e and automlability—reg
DATABAport two or
pplication, a can be chall
ew databaseo reverse en
nd stored prerstand the imencies (or in
managing coml model remd columns. Taintaining tr
As by:
ase design
nance time
ild a new one design. By will have oppe.
The H
mate the mangardless of th
ASE DEmore types company acenging.
e and its strungineer the docedures, asmpact of a pconsistencie
mmon objecoves all platThis provideaceability ba
ne, optimizinstaying in th
portunities t
High Perform
nagement ohe database
ESIGN of database
cquisition, or
ucture is to udatabase. Ds well as dep
potential chaes) among co
cts across datform-specifies traceabilitack to the ph
ng performanhe loop witho examine t
ance SQL Se
f these critic type or vers
e platforms. r the brainst
nderstand toing so allowpendencies. ange, generaommon data
atabases, esic propertiesty for businehysical
nce often sta data architehe physical
rver DBA
- 3 -
cal areas sion.
orm of
he data ws you By
ate a
pecially s and
ess
arts ects and data
Embarcadero
DATA NLogical datadesign is as Providing yorolled up/dooptimized fo
In the aboveand phone nkept some dCustomer tajoins when a Performancechange. In twith currentwhich tablesstrategies. Yspecific tabl
o Technologie
NORMALIa architects lflexible as p
our expertiseown, or horizor performa
e example, wnumbers fordenormalizaable with theaccessing th
e issues withthat case, mt row counts s may be goYou can alsoles.
es
IZATION ike to norm
possible. A fe with the phzontally/vertnce.
we have provr each customtion in the Ze Organizatie data.
h an existingmost data mo
and partitioood candidato get a clear
VS. PERalize the struflexible desihysical desigtically split) c
vided flexibmer rather t
Zip Code tabon Name an
g database modeling toolsoning propertes for new prer picture of
RFORMANucture so thagn is good,
gn on what tcan ensure t
ility by normhan limiting
ble with the Cnd Industry,
may make des allow you trties. Then ypartitioning,f table relati
The H
NCE
at data redubut not at th
tables can behat the desi
malizing a bitit to two an
City and Staallowing us t
enormalizingto reverse enyou can easil and you caonships and
High Perform
ndancy is rehe expense e denormalign is both fl
t to support d three. Hote Codes coto minimize
g the structungineer an ely create a ren rethink exi
d start strate
ance SQL Se
educed and tof performazed (i.e. merexible and
1 to N addreowever, we holumns and i
the number
re too drastexisting dataeport that reisting partitigizing how t
rver DBA
- 4 -
the ance. rged,
esses have in the r of
ic of a abase eveals oning to index
Embarcadero
OPTIM
CATCHIPerformancealways muchdevelopers database is pinpoint and Also, the roperformancrecognize acreate oppoidentify andwhile impro However, thor a differen
IDENTIFYWhen lookininto one of f
• I
• D
• A
• P
Understandthere are missue that, inhardware up So, how canthe databasFor exampleOracle or Mcurrent stateexecuting. By samplingperforming for CPU resoshown in theto free up th
o Technologie
MIZING
NG THE e problems h less expenhave been ra black boxd resolve pr
les of the dae DBA manand commun
ortunities to d fix poor pe
ving databa
he first step int type of bo
YING THng at databafour categor
nsufficient h
Database co
Application c
Poor-perform
ing the sourany cases wn the end, tupgrade fails
n we identifyse to calculate, all the ma
MASTER.SYSPe of the con
g these tableSQL, as welources and te first profilihe CPU. In th
es
G POOR
PROBLEcan occur at
nsive than letresponsible f, and they doblems.
atabase deveaging multip
nicate, how toffload SQLrforming codse performa
is to identifyottleneck.
HE MAIN ase bottleneries:
hardware
onfiguration
code proble
ming SQL
rce of the bohere a manaurns out to bto improve t
y the source te the load ojor databasePROCESS onections on
es, you can cl as where tithe highest Cng session ghis case, a h
R-PERF
EM EARLYt any stage otting them refor preventino not neces
eloper and dple platformsheir code is
L performancde before it ance.
y that the pe
SOURCEecks, it's imp
issue
em
ottleneck heager purchasbe a databasthe bottlene
of a performon the databe platforms hn SQL Servethe databas
create a cleame is beingCPU consumgraphic on pardware upg
FORMI
Y of the develoeach a prodng poor-persarily have a
database ads, you shouldaffecting dace tuning tohits produc
rformance p
E OF THEportant to cle
lps you allocses additionse configurateck.
mance bottlebase, pinpoihave perform
er and Sybasse and which
ar picture of spent in the
ming SQL stapage 7), thengrade is the
The H
NG SQ
opment lifecuction envir
rforming SQaccess to the
ministrator ad enable deatabase perfo developme
tion can sav
problem is re
E BOTTLearly identify
cate resourcal hardware tion or SQL
eneck? The knt poor-perfmance viewsse. These peh SQL statem
load on the e database. atement is on it’s not wor
best way to
High Perform
QL
cycle, but caonment. TraL code. For
e information
are expandinvelopers to
formance. Oent. Enablinge everyone t
elated to po
LENECK y the source,
es efficientlyto address tuning issue
key is to profforming SQLs, such as V$erformance vments the co
database anIf time is benly 3% of thrth the time allocate res
ance SQL Se
tching themaditionally, developers,n they need
ng. As a higsolve, or at
One strategyg developerstime and en
or-performin
, which usua
y. For exampa performan
e. The costly
file the activL, and top se$SESSION onviews describonnections a
nd identify ping spent wae total activito tune all y
sources.
rver DBA
- 5 -
m early is
, the to
h least is to s to ergy,
ng SQL
lly falls
ple, nce
vity of essions. n be the are
poor-aiting ity (as our SQL
The HHigh Performance SQL Server DBA
TV
t
TtT
Embarcadero
On the otheprofiling graClearly, this If the bottlewhat the issor if the appsolution to t
SQL TUOften timesoffending Shelp you ide
• A
• A
• D
• A
• I
For the finalstatement isreferred to atool) with thtwo table jo By drawing ratio, and fotable join siz
To the right isVisual SQL Tugood candidexecution pa
First we start of rows returnthe lowest, inat 0.34%.
Then down tothat join onlyThen we join ORDER_LINE115 rows, etc
o Technologie
er hand, if thaphic on pagwould be a
neck turns oue is so the
plication is dthe problem
UNING TEs, the main bQL has beenentify poor-p
Are the table
Are there an
Do any of th
Are all the un
s the databa
l step, of knos reasonableas Visual SQ
he detail taboin sizes.
the picture, ollow the joinze.
s an exampleuning diagramate for the beth is clear.
where the amned after filten this case OR
o CUSTOMER returns 8 rowup to
ES only givingc.
es
e worst perfge 7), you mmuch more
out to be accDBA can tak
doing a lot ofm is in the ha
ECHNIQUbottleneck isn identified, performing S
e and index
y missing in
e columns n
nique and n
ase choosing
owing whethe, the best a
QL Tuning or les hovering
you can findn to the next
e of a m. A est
mount ering is RDERS
RS as ws.
g us
forming SQLight spend a efficient ap
cess to a shake the approf single row nds of the d
UES a SQL statedevelopers
SQL:
statistics up
dexes?
need extend
ot-null const
g the right a
her the datand most effiVST. Try dra
g above the m
d the best pat table with t
L statement a day or longproach than
ared resourcopriate actiooperations (
developer.
ement takingcan analyze
to date?
ed histogram
traints corre
ccess path?
base chose cient step isawing out thmaster table
ath through the smallest
is consuminger trying ton upgrading
ce, the develon. However(when it sho
g up too mane the code. A
m or frequen
ectly defined
the right pats to lay the qhe tables (eites, indicating
the query. St result set si
g 86% of theo tune the SQ
the hardwar
oper needs , if the issueuld be using
ny resourcesAsk yourself
ncy statistics
on the colu
th, or even iquery out visther on your g table sizes
Start at the mze, then on
e CPU (see sQL statemenre.
to ask the D is row level
g arrays), the
s. Once the these quest
s?
umns and tab
f the SQL ually. This isown or usin, filter ratios
most selectivto the small
second nt.
DBA locking
e
ions to
bles?
often ng a s, and
ve filter est two
- 6 -
Embarcadero
In the profilinstatements.
However, in tthe load. The
o Technologie
ng session be
the profiling se load can be
es
low, a DBA m
session belowreduced sign
may spend too
w, we can see nificantly by tu
o much time t
that there is uning this one
The H
to make smal
one SQL state query.
High Perform
l gains by tryi
ement that ac
ance SQL Se
ing to tune in
ccounts for 86
rver DBA
- 7 -
ndividual
6% of
Embarcadero
STORAStorage maversions on to account fvolumes aredatabases, tOracle tablecommands The two maproblems camaximum catransaction capabilities allowing theincreasing dtransaction 100% guaraOracle tempdisk drive ru
A storage pglobally andspace in tabscan times.
New featureintroductionServer 2008data with a sotherwise rea starting powizards andspecific to sworking with
Oracle 10g database enIf hotspots aof global fraeliminating
In addition, must unders
o Technologie
AGE Mnagement isa variety of
for acceptabe properly mtransaction lespaces, temto manage t
jor storage man quickly brapacity or leprocessing sthat addres
eir underlyindata volumeslog files all hntee againstporary table uns out of sp
problem mad at the obbles and ind
es from seven of dynamic, SQL Serversolid mainteesulted in anoint for work other tools
support increh other data
introduced angine with a are discoveragmentationthe need fo
object storastand how to
es
MANAGs complex eoperating sy
ble file placemirrored and
ogs, and sempfiles, and the many div
managemenring down a
et a Microsofsystem reacts availabilityg operatings. IBM DB2 chave the capt failure. A Sspace temp
pace. Therefo
y also have bject level—dexes can a
eral databasec managemer DBAs have
enance plan,n interruptioking with youfor working
easing functabase platfor
automatic stset of disk ded, automat
n assistance, r full tablesp
age problemo detect and
GEMENnough, but wystems, the c
ement (isolatconfigured.gments in thredo logs inverse platfor
nt concerns adatabase. J
ft SQL Servets. Fortunate
y issues. All p system filescontainers, Opability to exSQL Server dpfile could bore, a DBA m
a major imcan result in
also cause p
e vendors caent views in Se additional v, you can pren of service.ur SQL Servewith SQL Se
ionality in thrms with SSM
torage manadrives for thetic relocationOracle8i an
pace reorgan
ms can created eliminate t
NT when you mchallenge beting log files) Today’s DB
he morning an the afternorms can be a
are always aust let an O
er transactionely, databasepopular datas to automatOracle datafxpand when database filee denied exmust still mo
pact on datn significant
problems an
an protect agSQL Server 2views to troueemptively d SQL Serveer databaseserver. The o
he database MS.
agement (ASe database tn is accompl
nd above offnizations.
e major perfthese issues
The H
anage differecomes extr) and ensureBA may be dand then ha
oon. Just rema difficult tas
vailability anracle archiven log fail to e vendors haabases now stically grow tfiles, or Micrnecessary. S could meet
xpansion at tonitor storag
tabase perfot unnecessad contribute
gainst these2005, and inubleshoot frdetect and rer’s Managems. There are
options presengine, and
SM), which ato automaticlished withoer locally ma
formance roaas quickly as
High Perform
rent databasreme. For SQe your operadealing with ve to switch
membering tsk.
nd performae log destinaexpand andave providedsport an “auto meet the osoft SQL SStill, such feat an imposedthe operatinge to ensure
ormance. Frary I/O for ae to excess
problems. Wncreased optagmentationesolve issuement Studio e substantial ented are, o
d there are n
allows a DBAcally manageut DBA inteanaged table
adblocks in s possible.
ance SQL Se
se platformsQL Server, yoating system
Sybase syst gears to de
the storage s
nce. Storageation reach see how a dd a number uto-extend” demand of erver databaatures are nod growth limg system levavailability.
ragmentatioa database. ive table an
With the tions with SQn. By couplis that could(“SSMS”) poptions inc
of course, veo provisions
A to provide e for file placrvention. In espaces,
a database.
rver DBA
- 8 -
s and ou need
ems, eal with syntax
e
dynamic of new feature
ase and ot a
mit or an vel if a
n—both Wasted
nd index
QL ing this have rovides luding rsion-s for
the cement. terms
A DBA
Embarcadero
Regardless
• Wastedspace thcontainenecessaarises w
• Poor Exobjects Ahead Mpages thpage pr
• Out of Sphysicallogical aback andmany timwhen de
• Forwardobject/tdatabaspage, orthe amoand in S
So, how doedatabases? no downtimshould estabthat allows ydatabases bmultiple typat least for c What aboutperformancplatforms. Tdiverse data The solutionthe followin
• Elimhete
• Allowdata
• Alertiden
o Technologie
of platform,
Space – Whan it needs ed in a data ry during tabith Oracle ta
xtent Proximare containe
Manager canhat are contioximity, sca
Sync Data O order of da
and physical d forth to obmes. In SQLetermining r
ded Rows – ext tables. He may either split the ro
ount of I/O nQL Server a
es a SQL SeAfter all, eac
me is attributblish an arouyou to configbefore you dpes of databacritical datab
t diagnosinge? This can b
The fact is thabase platfo
n is to desigg characteri
inates all theerogeneous
ws you to haabase platfor
ts you when tifying stora
es
object stora
hile it may ncan hamperwarehouse able scans. Tables that ha
mity – A dataed in extentsn read muchiguous. Syban times can
Order – Wheta stored in orders are i
btain the neeL Server, for oot cause fo
Each row inHowever, if ar relocate thw between t
necessary to nd Sybase a
rver DBA dech engine haed to space und-the-clocgure storage
develop storaases in their bases.
g and remedbe a challenere are vast rms.
n an advancstics:
e guessworkenvironmen
andle complrm.
storage issuage problem
age problem
not seem liker performancare not 100%
This increaseave high-wat
abase engines that are clo larger grouase is the saincrease.
en indexes hthe databasn sync, the deded informexample, in
or performan
a table shoa row expandhe row to anotwo or moreobtain the r
as forwarded
etect and diaas its own seoutages acr
ck, multi-plae thresholdsage-related environmen
ying databage for the Sdifferences
ced storage d
k and labor fts.
ex storage d
ues in the dams and pinpo
ms generally
e a problem,ce. For exam% utilized, ths response t
ter marks ab
e’s read-aheose to one aps of data wme. Convers
have a logicase, index accdisk head ca
mation. Othendex fragmence issues.
uld fit on a sds because oother page a
e pages. Neirow. This is sd rows.
agnose theseet of diagnosross many setform, fully-a and proactiproblems. S
nt and also f
se and objeSQL Server D
in diagnosti
diagnostic a
from handlin
dilemmas ev
atabase objeointing the o
The H
include four
, a table or imple, if the phe object wiltime for user
bove the area
ead mechaninother. For
when its targsely, when a
al order that cess performan scan in onrwise, the dintation is on
single page,of the growtand leave bether situatio
seen in Orac
e object stostics. Moreoervers? As a automated dive notificatiSuch a plan sfocus on bein
ct storage pDBA tasked wic and treatm
and manage
ng database
ven if you are
ects are draibjects that r
High Perform
r componen
ndex with mpages of a SQl examine mr queries. Thas where dat
isms work mexample, SQet object han object has
does not mmance can bene direction isk head willne of the mo
if possible, th of variableehind a poin
on is preferrecle as chaine
rage probleover, how ca
high performdatabase moons across ashould take ng complete
problems thawith taking oment metho
ment system
storage pro
e unfamiliar
ning performrequire atten
ance SQL Se
nts:
more allocateQL Server o
more pages the same prota resides.
more efficientQL Server’s Ras extents ans poor exten
atch the acte affected. Ifinstead of m skip across
ore common
except for lae length datnter to the need as they ind or migrate
ms in less fan you ensuremance DBAonitoring soall your criticinto accounely automate
at hurt on additionads that exist
m that incorp
oblems in
with the und
mance by visntion.
rver DBA
- 9 -
ed bject than blem
tly when Read nd nt or
ual f the
moving the disk culprits
arge ta, the ew crease
ed rows
amiliar e that , you lution
cal t the ed 24/7,
al t across
porates
derlying
sually
Embarcadero
Because stoyou must betwo-phase aspace issuesneeds of yohigh growth
A best practexample, crobjects that
Once these again.
PERFODBAs want tuses a systeshortest posdatabase plto the next. the databas
The startingand then apmethod. Th
• Bott
• Wor
• Ratio
The followin
BOTTLE
Regardless waiting to pdatabase boperceived b
Bottleneck awhere a datnumber of tactivity coun
If Oracle latperformancpotential socausing the
o Technologie
orage manage proactive iapproach to s threateningur environm
h rates so yo
tice is to rebeate a find-at violate you
jobs are set
ORMANtheir databam that connssible respoatform. As wAs a result,
se engines yo
g point is to dpply a set of ese method
leneck/Resp
kload Analys
o Analysis
ng examines
NECK/Rof the platfo
perform workottleneck ex
bottlenecks a
analysis is antabase has btimes specifinters will hel
ch contentioe, you can u
ources has beproblem.
es
gement can n your approsetting thre
g overall sysment. For SQ
u have plent
build your daand-fix job tr specified s
t up and sch
NCE Mases to run anects to thosnse times. A
with storage you need toou manage.
define the ddiagnostics
ds of analysis
ponse Time A
sis
s how each m
RESPONS
orm, when a k. A process xists. DBAs uare hurting p
n essential mbeen spendin
c queries arlp point you
on or heavy use bottleneceen identifie
be complexoach and au
eshold alertsstem performL Server, tarty of notice
atabase objehat searches
set of thresho
eduled, you
MANAGs quickly as
se databasesAgain, the sit
managemeo put togeth
different metand actions
s include:
Analysis
method can
E TIME Adatabase is that is waitise bottlenec
performance
method of mng its time. Ie run and hoin the direc
DB2 table-scck analysis toed, drill dow
x, especially utomate as m starts with s
mance, then rget your datto solve pot
ects for maxis the entire dolds. Then r
u will never h
GEMENpossible. Ths—they wantuation is cont, the variaer a platform
thods of anas that can be
be used to m
ANALYSI
running, evng may meack or respon
e.
easuring pen SQL Serve
ow long theyction of a poo
can activity ho pinpoint th
wn to detail a
The H
if you work omuch of the standard thrcustomizesta and log fitential proble
imum perfordatabase (oreorganize th
have to manu
T he same cant their querimplicated wbles and rem
m-neutral ro
lysis that wile used for ea
manage the
IS ery connecte
an nothing, onse time ana
erformance ber, looking ay run, along orly perform
has been drahe root caus
about the ses
High Perform
on multiple dmanagemenesholds desthem to me
ile groups thems.
rmance at rer a particularhem for bett
ually reorgan
be said for es and proc
when you havmedies diffeadmap that
ll be used acach platform
performanc
ed process ior it can be alysis to dete
because it heat Buffer Cac
with disk quming query o
agging a dase. Once onessions and o
ance SQL Se
database plnt as possibligned to un
eet the uniquhat are expe
egular intervr set of objeter performa
nize your da
every persoesses to havve more thar from one papplies to e
cross all plat under each
ce of any dat
is either woran indicator ermine wheth
elps you tracche Hit Ratioueues and otr hardware i
tabase’s e or more of
objects that a
rver DBA
- 10 -
atforms, e. A cover
ue riencing
als. For cts) for
ance.
tabases
n who ve the n one platform each of
tforms
tabase.
rking or that a her
ck os, ther issue.
f the are
Embarcadero
This methodthe case, nebottlenecksenable DBASQL Server SQLPERF(WSybase offe
WORKLO
Workload a
• Sess
• SQL
When perfoworkload. Tthe bulk of rare run safeDBA should
Normally, thconsumptioa session’s cdynamic ma
Understandthe second will producephysical desSQL? Naturmore subtlewhether sorbecomes m
All the datametrics offecollect and SQL execut
RATIO ARatio-baseddatabase sloprocedure pthat such ratanalysis. Thethe Oracle l
When workiutilization osee whether
o Technologie
dology is theearly every d and respon
As to understmade wait e
WAITSTATS) rs new moni
OAD AN
nalysis involv
sion resource
L execution a
ormance on aThe issue is sresources foly in non-pe
d examine ea
his can be eaon and statistcurrent and anagement v
ing current aset of data p
e the secondsign being thally, factors
e factors canrts are done ore of an art
base vendorered by any vmeasure SQion views in
ANALYSIS
d analysis waowdown. Yoplan hit ratiotios are nowere is an eleibrary cache
ng through n the serverr and how of
es
e strategy ofatabase ven
nse times. Ortand bottlenevents availaand now exitoring views
NALYSIS
ves the inve
e consumpti
analysis
a database dystem balanr an extendeak hours. If i
ach problem
asily accomptical executihistorical actviews, one b
and historicapoints needed-best perfohe first. But wsuch as over make a diffin memory ot form than a
rs offer a winvendor. Micr
QL and storeversions 12.
S as for many you are probao, Sybase datw worthless a
ment of trute hit ratio an
performanc. In additionften indexes
f choice for tndor has tailoracle 10g intnecks and reable via an uposes that ds in engine v
stigation of
on and activ
drops suddence. In a welled period ofindividual se
m session in d
plished by vion statisticstivities. SQL
being dm_ex
al SQL execued to propermance-enhwhat set of mrall elapsed erence—likeor on disk. Aa science.
ndow into SQrosoft SQL Sd procedure5.03 and ab
years the onbly all too fata cache hit
and misleadith to these cd the Micros
ce issues, bufn, by lookings are used an
top performor edits engtroduced nesponse timendocumente
data via dynaversions 12.5
two critical a
vity
enly, one or tl-balanced sf time, with tessions are udetail to unc
ewing sessio. Most datab
L Server, for ixec_session.
ution patterrly perform wancing boosmetrics shoutime, CPU t
e the numbeAs we saw on
QL analysis. Server offers e executionsove. DB2 of
ly techniqueamiliar with tratio, and song because
claims, but sosoft and Syb
ffer cache hig into the qund whether t
The H
ance analystines to repow metrics in
es at global ed DBCC coamic manag5.03 and abo
areas of data
two sessionsystem, no sithe exceptiousing a majocover session
on metadatabase engineinstance, pro
ns will enablworkload anst available fuld you use time, and I/O
er of times a n page 6, thi
Oracle 10g code profili
s. Sybase hasffers its own
e DBAs had tthe Oracle bo forth. Man of advancesome ratios a
base procedu
it ratios in SQueries and qthe indexes
High Perform
ts in the induort metrics to
its V$ perfoand session
ommand (DBement view
ove.
abase perfo
s often genengle session
on of batch jrity of systemn activity.
a coupled ws provide vaovides this d
le a databasnalysis. Optimfor a databato evaluate ‘
O activity plastatement i
is is where S
has the mosng utilities ts made a goSQL tracing
to diagnose buffer cache ny performans in bottlene
are still quiteure plan hit
QL Server wuery executihave been c
ance SQL Se
ustry. This bo analyze ormance view
levels. MicroBCC
ws (“DMVs”),
rmance:
erate the buln should conob processem resources
ith resource ast details redata via seve
se analyst to mizing SQL se—with pro‘good’ versu
ay a part. Buts executed, QL analysis
st complete that can be uood start withg capabilities
the cause ohit ratio, Mi
nce experts aeck/responsee valuable, suratios.
will help definion plans, yocreated on t
rver DBA
- 11 -
eing
ws that osoft
and
lk of the nsume es that , the
egarding eral
have code oper us ‘bad’ t other, or
set of used to h its s.
of a crosoft advise e time uch as
ne ou can he most
Embarcadero
beneficial coindexes), yoto determin
Ratios can bfeature setsreporting. Fissues of dacause of pe
More obscuforwarded runless you kSybase tabletypes of situ
Similar meato address. level. Thesesubsystem tindex analys
CAPACMany DBAsday activitieproper capasuch as, “Hocurrently un
Capacity pla
• Colle
• Histo
• Fore
Every organinformation
• SnapstatisElemwell
• Snapwait
• Snapand
While someand performtasks are co
o Technologie
olumns in thou can gain i
e which inde
be helpful, b, server funcFor exampleta access, mrformance is
ure ratios canows, would know how me were forwa
uations, the a
sures exist f There are a
e counters wto the instansis, you can
CITY M find it diffic
es simply leaacity planninow much moderutilized?
anning gene
ection of key
orical analys
ecasts of futu
nization will h:
pshots of glostics, Oracle
ments such aas other per
pshots of peevents
pshots of resuser traffic
e DBAs have mance, the b
mplicated g
es
he queried tansights into exes may ne
ut the informctionalities ae, the use of making it diffssues.
n also be useyou conside
many total rowarded, this isapplication o
or SQL Serva host of couill let you exce performagain additio
MANAGcult or almosve no room
ng can help bore storage w”
erally involve
y database m
is of collecte
ure needs
have differen
obal storagee tablespaceas total allocarformance-re
erformance m
source consu
been able tburden in usigreatly by mu
ables. By loolikely index
eed to be reb
mation is bend other issa SAN disk icult to dete
eful. For exaer this good ws are in thes enough infof proper ra
er and will bunters availabxplore and reance. Furtheonal insights
GEMENt impossiblefor historica
both the DBwill this data
es three proc
metrics
ed metrics
nt needs, bu
e usage, inclue data, and dated space, elated items
metrics, inclu
umption, inc
to get by using such teculti-platform
oking at DMcandidates.built or reor
ing muddiedues outside subsystem c
ermine wheth
ample, if youor bad? Of c
e table. But iformation totio analysis c
be helpful bable from Perecord informer, with the Dinto sugges
NT e to implemeal and proactBA and manaabase need i
cesses:
ut in general
uding Sybasdatabase obj
used space,s like numbe
uding key da
cluding metr
ng scripts anhniques for
m database e
The H
MVs for sugg Then you cganized.
d in many inthe databas
can confuse her a slow d
u were told tcourse, you if you knew t
o make a reocan be adva
ased on the rformance M
mation for evDMVs availastions for inv
ent capacity tive analysisagement, anin six month
, most shoul
se device meject statistics, and free sp
er of chained
atabase stati
rics such as C
nd manual pcapacity pla
environments
High Perform
ested indexecan review fr
stallations, bse engine’s rdisk access isk system m
hat a Sybasecannot answthat 98% of
organization ntageous.
specific situMonitor at th
erything froble for quer
vestigations.
manageme. This is unfoswering imps?” and “Is t
ld collect the
etrics, SQL Ss that conce
pace should d rows
istics like I/O
CPU usage,
processes tonning is usus, as well as
ance SQL Se
es (and unusragmentatio
by factors surealm of conreporting to
may be the ro
e table had 1wer this quesall rows in yodecision. In
ations you’ree operating m the disk y performan
ent, as their dortunate becportant questhis databas
e following
Server databern space usa
all be collec
O performan
memory usa
manage stoually too high
the need to
rver DBA
- 12 -
sed n ratios
uch as ntrol and o mask oot
100 stion our these
e trying system
nce and
day-to-cause stions se server
ase age. cted as
ce and
age,
orage h. These build
The HHigh Performance SQL Server DBA
Embarcadero
and maintaichallenges iwell as analythese compSQL Server
The followin
• Build
• Crea
• Sche
• Implperfoform
o Technologie
n a repositoin the time ityze and man
plications, it iDBA—espe
ng technique
ding a repos
ating statistic
eduling to au
ementing toormance, an
matted HTML
es
ory that holdt takes to manage the genis obvious thcially when
es to simplify
sitory to trac
cal collectio
utomatically
ools to accurnd resource L reports tha
s all the necanually weavneration of chat manual mmoving into
y capacity m
ck key databa
ns for everyt
run collectio
rately predicrequirement
at managers
cessary statisve together complex histmaintenance other DBM
management
ase metrics
thing you wa
ons as often
ct future neets through veasily unde
stics. In additcollection sctorical and fo
e is not practS platforms.
t will save yo
ant to track a
n as needed
eds by depicisual graphsrstand
tion, DBAs acripts and scorecasting rtical for the h
ou time and
across your k
cting upcoms (example b
are faced witcheduled jobeports. Withhigh perform
headaches:
key databas
ing storage,below) or we
th bs as h all mance
es
, ll-
- 13 -
Embarcadero
CHANToday’s SQmodelers, dgiven time, of which maparticular dainclude a dachange man Very few orgassume thatmanagemenrobust datawith peculia As the DBAthe changesbaselines anperiodically the baselineplatform, prdatabase cocould have a Third party acontrol, at tcomprehenschanges thapoint to diahave to rely Just as impocompliance Being able tchanges, ca Without a roputting the serious dataadminister t
CONCThe combinand lower hmaintain higimplement a
o Technologie
NGE MAL Server DBA
database deveach databa
ay contain anatabase is tyata modelingnagement to
ganizations lt database cnt, software base change
arities that m
, you shoulds occurring and manage crevisit the b
e. Having a rovides confonfiguration as much imp
applicationshe end of thsive databasat occur whegnose probsolely on th
ortant as objand reporti
to flag erronn be critical
obust changwork of the
abase issue othe database
CLUSIOnation of mueadcount ha
gh databasea strategy fo
es
ANAGA sits at the velopers, arcase is physicny one of sevypically storeg tool, a SQool, etc.
look at datachange manaconfiguratio
e managememake it quite
d develop a dacross all of change to th
baseline to econsistent vidence that into your ch
pact as an ob
s present thehe day, you ase change men vendor uplems that ma
he quality of
ject level chng pressure
neous, or eveto meeting
ge managemother discipor, even wore.
ON ltiple databaas become a availability
or ensuring g
EMENTcenter of a
chitects, buscally instantiaveral versioned in severalL developm
base changeagement is aon management solution different fro
database chyour multipl
hose baselinensure that eiew with condatabase ch
hange managbject change
eir own uniqare still respo
managementpdates are day arise whethe vendor’
anges are chs, DBAs musen maliciousexpected co
ment solutionplines of admrse, could be
ase types ana real challenand perform
good design
T complex sys
siness analysated across ans of the sam locations usent tool, a d
e managemea natural out
ment, and dais to recogn
om software
ange manage database es. In addit
elements arensistent repohange is beingement proe.
ue set of chaonsible for at process allo
deployed. Wen applying ps updates an
hanges madst establish ts changes toompliance le
n for your muministration ae a direct ref
nd versions, ange for the S
mance. The kn documenta
The H
stem that tousts, software a number of
me databasesing a multip
database adm
ent as a singtcome, at thata modelingnize it as a stchange man
gement procplatforms. Yion, best pra
e properly veorts, no mattng managedcess. A cha
allenges. Than applicatioows you to u
With that infopatches. Witnd their abil
de to the secthat they hav
o database seevels.
ultiple databat risk; a singflection on y
added databSQL Server Dkey to keepination and hig
High Perform
uches many developers
f different ene. Furthermoplicity of tooministration
gle, unified pe juncture o
g. The first sttandalone pnagement.
cess that proYou should eactices dictaersioned andter the databd. You can ange to a con
hough most on’s performunderstand aormation, thethout a procity to remote
curity modelve control oecurity, and
base platformgle change cyour ability to
base compleDBA who wong databasegh quality SQ
ance SQL Se
participants, and more.nvironmentsore, the desigols, which ma
tool, a data
process. Manof database tep in develrocess or dis
ovides a viewestablish sol
ate that you d if needed, base also incorponfiguration s
limit the DBance. A and predict e DBA has a cess in placeely troublesh
. With increver the datathen act on
ms, you coulcould result o successful
exity, more dorks hard to es running wQL. Proper s
rver DBA
- 14 -
s: data At any , each gn of a ay base
ny
oping a scipline,
w into lid
reset
rate setting
BA’s
starting , you hoot.
asing abase.
those
d be in a ly
data,
well is to storage,
Embarcadero
performancshould be p In the end, tDBA. And aavailability a
ABOUEmbarcaderperformancchange manat hand. Emdeliver prodenables youtools will givensuring the
Another vaGroup, SSWmany otherSSWUG.ORfrom expertthat address
ABOU
o Technologie
e, capacity mplatform-inde
the responsi well-prepar
and perform
T EMBro Technoloe SQL DBAnagement. Embarcadero ductivity gau to simplifyve you the ce success of
luable resouWUG.ORG, wr resources tRG newslettets. The site s very specif
T THE
es
managemenependent an
ibility for thered DBA shoance.
BARCAgies provide
A to take coER/Studio al
DB Optimiins, regardl
y, automate, confidence to
your databa
urce for higwhich provito help youer to stay onallows you
fic challenge
AUTH
Scodevfor direon joinleadcom
SteSSW198wee12 bspothe
nt, and changnd as autom
e performancould have all
DERO Aes a completontrol of SQlows you tozer XE, DBAess of your and report
o carry out eases.
gh performades articles manage th
n top of lateto search th
es that affect
ORS ott Walz hasvelopment aEmbarcade
ection of thedatabase de
ning Embarcd for Louisvimputer infor
phen WynkWUG.ORG. H9 and writes
ekly SQL Serbooks transla
oken at numehappenings
ge managematic as poss
ce of an org the tools sh
AND Ste set of pro
QL, storage, reverse engArtisan XE,
current levdatabase c
even the mo
ance SQL D, video pro
hese compleest advancemhe site for spt your organ
s more than and serves asro Technolo
e company'sevelopment cadero four yille Gas & Elmation syste
oop is a MicHe has works a daily colurver video prated into 7 lerous technos in the tech
The H
ment are alsoible.
ganization’s dhe needs to
SSWUofessional-gr
performancgineer the Eand Analyst
vel of experhange. In ad
ost complex
DBAs is thegrams, web
ex environmments, discopecific articleization’s dat
15 years of es senior dire
ogies. In this database pand adminis
years ago, Sectric. He hoems from W
crosoft SQL ed with SQL
umn and newrogram. Stepanguages aology confernology worl
High Perform
o critical. Su
database strensure high
G.ORGrade databasce, capacity
ERD so you ut Series Toortise. DB Chddition, thestasks, which
SQL Servebcasts, virtua
ments. You over valuables, editorialtabase envir
experience iector of prod
position, Scroduct famistration prodcott served olds a bacheestern Kentu
Server MVPL Server sincwsletter on tphen is the around the wrences and rd.
ance SQL Se
ch a strateg
ructure falls o database
G se tools for t
y managemeunderstand ols are desighange Manase essential
h goes a long
er Worldwidal conferenccan registere insight, ans, scripts, anonment.
n database duct managecott overseesly, while focuducts. Prior tas developm
elor's degreeucky Univers
and the foue its inceptiohe site, as wauthor of mo
world. Stephereports regu
rver DBA
- 15 -
y
on the
the high ent, and the task gned to ager XE support
g way to
de Users ces, and r for the nd learn nd more
ement s the using to
ment e in sity.
under of on in
well as a ore than en has ularly on
Embarcadero Technologies, Inc. is the leading provider of software tools that empower application developers and data management professionals to design, build, and run applications and databases more efficiently in heterogeneous IT environments. More than 90 of the Fortune 100 plus an active community of more than three million users worldwide rely on Embarcadero’s award-winning products to optimize costs, streamline compliance, and accelerate development and innovation. Founded in 1993, Embarcadero is headquartered in San Francisco with offices located around the world. Embarcadero is online at www.embarcadero.com.