Sie sind auf Seite 1von 24

vikrom Sinqb Soini

ISAS
lnLel v1une ls sofLware from
lnLel CorporaLlon lL ls been
used by developmenL
companles Lo analyze Lhelr
sofLwares and appllcaLlons
performance
Along lL also provldes advlce
Lo boosL performance of apps
on speclflc CS"
9age of


Acknowledgment
l am graLeful Lo my parenLs relaLlves and frlends who help us ln documenLaLlon of lnLel
v1une lSAS
l am also graLeful Lo our faculLy Mr kam Sharma (GL NII1 A8C) for hls keen guldance and
supporL whlle documenLaLlon of Lhls documenL ln your hands l also wanL Lo Lhank Cod who
glves knowledge and poLenLlallLy Lo wrlLe Lhls all documenLs clearly and presenL lL ln fronL of
you readers
uurlng Lhe documenLaLlon l feel loL of Lroubles such as crashlng of Wlndows and damaged
Lo compuLer hardware hlndrance ln worklng of sofLware lnLel v1une l would also llke Lo
menLlon Lhe name of Lhe person our frlend Sandeep Chaudhary for provldlng hls 9C for
presenLlng Lhls appllcaLlon
And aL lasL l would llke Lo say a speclal Lhanks Lo nll1 who provldes us a wonderful chance Lo
presenL Lhls documenL ln fronL of you all readers 1hanks everybody who dlrecLly or
lndlrecLly helps for presenLlng Lhls flle
vlkram Slngh Salnl


9age 3 of

ontents at Clance

1 lnLroducLlon 4
2 Sampllng
21 lnLroducLlon7
22 Sampllng mechanlsm7
23 ulffbeLween 18S L8S8
24 WhaL happens durlng Sampllng8
23 leaLures of Sampllng9
26 Sampllng Cver 1lme10
3 Call Craph
31 lnLroducLlon12
32 leaLures of Call Craph12
4 CounLer MonlLor
41 lnLroducLlon16
42 leaLures of CounLer MonlLor16
43 Worklng of CounLer MonlLor19
3 1unlng AsslsLanL
31 1unlng AsslsLanL20
32 1unlng AsslsLanL ConcepLs20
33 leaLures of 1unlng AsslsLanL20
34 undersLandlng 1unlng MeLhodology21
33 SLraLegles for lmprovlng 9erformance21
36 1ypes of Advlce21
37 lnformaLlon LhaL 1unlng AsslsLanL provldes22
6 8eferences24







9age of

Introduction
1be v1ooe ooolyzet ptovlJes oo loteqtoteJ petfotmooce ooolysls ooJ tooloq eovltoomeot
tbot belps yoo ooolyze yoot coJes petfotmooce oo systems wltb lAJ2 lotel(k) 64 ooJ lA64
otcbltectote
v1une analyzer can plug ln lnLo MlcrosofL vlsual SLudlo and Lcllpse lnLegraLed developmenL
envlronmenLs
Cne can work wlLh Lhe v1une analyzer uslng Lhe graphlcal lnLerface and command llne
lnLerface All commands Lo creaLe and run AcLlvlLles musL be preceded by vt| aL Lhe
command llne
LINUk SUCk1
1he v1une(1M) 9erformance Analyzer can analyze Lhe performance of your Llnux*
appllcaLlon 1he v1une analyzer ls lnsLalled on a conLrolllng sysLem and conLrols Lhe run of
your Llnux appllcaLlon on a 8emoLe AgenL sysLem 1he v1une analyzer Lhen collecLs daLa on
your Llnux appllcaLlon by collecLlng daLa remoLely
IAVA SUCk1
When Lhe v1une(1M) 9erformance Analyzer analyzes Lhe performance of your !ava*
appllcaLlon or appleL (class) Lhe vlrLual Machlne (vM) and !usLln1lme Compller (!l1) are
enhanced Lo provlde Lhe v1une analyzer wlLh speclflc lnformaLlon requlred Lo analyze Lhe
performance of a !ava appllcaLlon
uurlng sampllng Lhe vM and !l1 provlde Lhe v1une analyzer wlLh lnformaLlon abouL !l1
complled !ava meLhods belng loaded lnLo memory such as Lhelr memory addresses slzes
and symbol lnformaLlon
NL1 SUCk1
1he v1une(1M) 9erformance Analyzer enables you Lo proflle nL1* and AS9nL1 web
servlces runnlng on your machlne
1he v1une analyzer wlll seL Lhe necessary envlronmenL varlables and resLarL Lhe web servlce
before collecLlng sampllng or call graph daLa 1he envlronmenL varlables wlll be deleLed and
Lhe servlce resLarLed on compleLlng daLa collecLlon
use Lhe somplloq cooflqototloo wlzotJ and coll qtopb cooflqototloo wlzotJ for proflllng
AS9nL1/nL1 web servlces

9age of

LA1ULkS C IN1LL V1UNL LkCkMANCL ANAL2Lk
1 CALL GkAn 9rovldes a graphlcal vlew of Lhe appllcaLlon and helps you ldenLlfy
crlLlcal funcLlons and Llmlng deLalls ln Lhe appllcaLlon
2 SAMLING CalculaLes Lhe acLual performance of an appllcaLlon over a perlod
(1lmebased sampllng) and for varlous processor evenLs(LvenLbased sampllng)
3 CCUN1Lk MCNI1Ck 9rovldes sysLem level performance such as resource
consumpLlon durlng Lhe execuLlon of an appllcaLlon
4 1UNING ASSIS1AN1 9rovldes Lunlng advlce from an anzlusls of Lhe performance
daLa 1he Lunlng advlce helps you lmprove performance of an appllcaLlon
3 nC1SC1S VILW elps ldenLlfy Lhe area of code LhaL Lakes Lhe maxlmum C9u
Llme
MINIMUM kLUIkLMLN1S C SC1WAkL
1 nAkDWAkL kLUIkLMLN1S
Processors 5upported
Servers
O ;uadCore lnLel(8) xeon(8) 9rocessor 3300 Serles
O uualCore lnLel(8) xeon(8) 9rocessor 3100 Serles
O uualCore lnLel(8) xeon(8) 9rocessor 3000 Sequence
O uualCore lnLel(8) xeon(8) 9rocessor 7100 Serles
O uualCore lnLel(8) xeon(8) 9rocessor 7000 Sequence
O uualCore lnLel(8) xeon 9rocessor Lv
O lnLel(8) xeon(8) processor M9
O lnLel(8) xeon(8) processor
O uualCore lnLel(8) lLanlum(8) 2 processor 9000 sequence
O Low volLage lnLel(8) lLanlum(8) 2 9rocessor
O lnLel(8) lLanlum(8) 2 processor
Desktop
O lnLel(8) Core(1M)2 ;uad processor
O lnLel(8) Core(1M)2 LxLreme processor
O lnLel(8) Core(1M)2 uuo processor
O lnLel(8) Core(1M) uuo processor
O lnLel(8) Core(1M) Solo processor
O lnLel(8) 9enLlum(8) u processor 900 sequence
O lnLel(8) 9enLlum(8) u processor
O lnLel(8) 9enLlum(8) 4 processor LxLreme LdlLlon
O lnLel(8) 9enLlum(8) processor LxLreme LdlLlon
O lnLel(8) 9enLlum(8) 4 processor
Mob||e
O Moblle lnLel(8) 9enLlum(8) 4 9rocessor M
O lnLel(8) 9enLlum(8) M processor
9age of

O lnLel(8) Celeron(8) M processor
O lnLel(8) Celeron(8) u processor
O lnLel(8) Celeron(8) processor
O Moblle lnLel(8) Celeron processor

SC1WAkL kLUIkLMLN1S
3b|t operat|ng systems support|ng IA3 processors
O MlcrosofL* Wlndows x9 9rofesslonal Servlce 9ack 2
O MlcrosofL* Wlndows Server 2003 LnLerprlse LdlLlon Servlce 9ack 1
O MlcrosofL* Wlndows Server 2003 82 LnLerprlse LdlLlon
O MlcrosofL* Wlndows vlsLa*
O MlcrosofL* Wlndows Server 2008 8C0 (bulld 6001)
b|t operat|ng systems support|ng Inte|(k) processors w|th Inte|(k) LM1
O MlcrosofL* Wlndows x9 9rofesslonal x64 LdlLlon
O MlcrosofL* Wlndows Server 2003 LnLerprlse x64 LdlLlon
O MlcrosofL* Wlndows Server 2003 82 LnLerprlse x64 LdlLlon
O MlcrosofL* Wlndows vlsLa*
O MlcrosofL* Wlndows Server 2008 8C0 (bulld 6001)
b|t operat|ng systems support|ng Inte|(k) Itan|um(k) arch|tecture processors
O MlcrosofL* Wlndows Server 2003 LnLerprlse LdlLlon Servlce 9ack 1
O MlcrosofL* Wlndows Server 2008 8C0 (bulld 6001)
3 SS1LM MLMCk kLUIkLMLN1S
AL leasL 128 MegabyLes of 8AM
DISk SACL kLUIkLMLN1S
O AL leasL 103 MegabyLes of avallable space on a local drlve
O 20 MegabyLes of dlsk space ls requlred for sysLem flles on Lhe drlve conLalnlng Lhe sysLem
dlrecLory (for example C)
1he addlLlonal hard dlsk space ls needed for updaLlng and lnsLalllng Lhe uLLs and CCxs LhaL Lhe v1une
analyzer requlres Lo be ln Lhe sysLem dlrecLory





9age of

Sampling
IN1kCDUC1ICN
Sampllng ls Lhe process of collecLlng a seL of daLa for analysls and represenLlng Lhe analyzed
daLa ln a sLaLlsLlcal formaL use Lhe collecLed daLa Lo ldenLlfy Lhe crlLlcal processes Lhreads
modules funcLlons and llnes of code runnlng on sysLem
uurlng sampllng Lhe v1une(1M) 9erformance Analyzer monlLors all Lhe sofLware execuLlng
on your sysLem lncludlng Lhe operaLlng sysLem !l1complled !ava* appllcaLlons nL1*
appllcaLlons and devlce drlvers
Sampllng does noL modlfy blnary flles or execuLables ln order Lo monlLor Lhe performance
of appllcaLlon 1he v1une analyzer analyzes Lhe collecLed samples and helps you Lo ldenLlfy
otspots ls a secLlon of code wlLhln a module LhaL Look long Llme Lo
execuLe 1hls resulLs ln hlgh amounL of processor Llme spenL execuLlng LhaL
secLlon Lhus generaLlng loL of samples for LhaL module
2 ott/eNecks ls an area ln Lhe code LhaL ls slowlng down Lhe execuLlon of
appllcaLlon 8oLLlenecks appears as hoLspoLs ln hoLspoL vlew 8emovlng
boLLlenecks and hoLspoLs opLlmlze Lhe appllcaLlon
1WC 1LS C SAMLING MLCnANISM 1C CCLLLC1 SAMLING DA1A
1 1IML8ASLD SAMLING (18S) 1he v1une(1M) uses Lhe operaLlng sysLem Llmer
Lo lnLerrupL and collecL samples of all acLlve lnsLrucLlon addresses aL a regular Llme
lnLerval (1ms by defaulL) 1he collecLed samples provlde Lhe performance daLa of all
Lhe processes runnlng on Lhe sysLem 9rocesses LhaL Look Lhe longesL Llme Lo
execuLe have Lhe hlghesL number of samples
LVLN18ASLD SAMLING (L8S) use Lo ldenLlfy sysLemwlde sofLware
performance problems caused by processor evenLs such as Cache Mlsses and
MlspredlcLed 8ranches
lrom Lhe L8S daLa one can deLermlne whlch process Lhread module funcLlon and
source llne ln program generaLed Lhe mosL processor evenLs and lf any of Lhose
evenLs lmpacLed Lhe performance of program 1he v1une analyzer provldes
predeflned event rotios recommended for use by performance analysLs aL lnLel
9age of


I16u8L : Lvent based samp1ng
DILkLNCLS 8]W 18S AND L8S
5 uaLa ls collecLed uslng ClockLlck evenLs 8uL when L1 lnsLrucLlons are
execuLed by processor clock Lhe processor clock causes Lhe clockLlcks evenLs Lo
sLop occurrlng 1hls resulLs ln no samples belng collecLed whlle Lhe processor ls ln
halL sLaLe 1he v1une wlll reporL few samples Lhan you were expecLed
15 uaLa ls collecLed uslng CS Llmer And CS Llmer ls noL affecLed durlng L1
lnsLrucLlons And Lhe samples are collecLed accuraLelLy 18S can poLenLlally glves
more accuraLe daLa
WnA1 nALNS DUkING SAMLING
When you run an AcLlvlLy conflgured wlLh Lhe sampllng collecLor Lhe v1une analyzer does
Lhe followlng
O WalLs for Lhe delay sampllng Llme (lf speclfled) Lo elapse and Lhen sLarLs collecLlng
samples
9age of

O lnLerrupLs Lhe processor aL Lhe speclfled sampllng lnLerval and collecLs samples of
lnsLrucLlon addresses lor every lnLerrupL Lhe v1une analyzer collecLs one sample
O SLores Lhe execuLlon conLexL of Lhe sofLware currenLly execuLlng on sysLem
LA1UkLS C SAMLING
1he followlng are Lhe maln feaLures of Lhe sampllng collecLor and vlews
1 Co||ect|on
O Mu|t|p|e event samp||ng 9erform evenL based sampllng wlLh mulLlple evenLs ln one
run uependlng on Lhe Lype of processor uslng Lhe v1une analyzer can monlLor and
collecL samples on Lwo or more evenLs ln one run
O kemote samp||ng CollecL sampllng daLa for an appllcaLlon runnlng on a remoLe
sysLem ?our remoLe sysLem can be a machlne runnlng on any operaLlng sysLem
supporLed by Lhe v1une analyzer
O CollecL sampllng daLa for appllcaLlons runnlng on sysLems enabled wlLh nyper
1hread|ng 1echno|ogy
V|ews
1he followlng sampllng vlews help you analyze Lhe daLa
O 1hread v|ew vlew Lhe Lhreads runnlng wlLhln a process and selecL one or more
Lhreads Lo drlll down Lo speclflc hoLspoLs
O Summary v|ew Cpens defaulL for clockLlck evenLs
O rocess v|ew ulsplay a sysLemwlde vlew of all Lhe processes runnlng on your
sysLem when sampllng daLa was collecLed
O Modu|e v|ew ulsplay all Lhe modules wlLhln selecLed Lhreads
O notspot v|ew ulsplay funcLlon names assoclaLed wlLh selecLed modules Croup
hoLspoLs by funcLlon relaLed vlrLual address (8vA) source flle or class
3 Accessor|es
1he followlng panels and Loolbar opLlons are avallable from Lhe sampllng vlew
O Samp||ng too|bar A sampllng Loolbar ls avallable aL Lhe Lop of each sampllng vlew
1hls Loolbar lncludes buLLons labeled 9rocess 1hread Module oLspoL and Source
SelecL lLems wlLhln a vlew and cllck one of Lhe buLLons Lo drlll down
O 1abbed w|ndows When you open a speclflc sampllng vlew a Lab ls creaLed aL Lhe
boLLom of Lhe wlndow labeled wlLh Lhe name of Lhe vlew for example 9rocess
1hread Module or oLspoL lf you open several vlews a Lab for each open vlew ls
creaLed aL Lhe boLLom of Lhe wlndow ?ou can use Lhe Labs Lo qulckly move from one
vlew Lo anoLher
O M|crosoft Lxce| ulsplay your sampllng daLa ln a MlcrosofL Lxcel 2000 spreadsheeL
?ou can cusLomlze Lhe appearance of Lhe spreadsheeL reporL as needed
9age 10 of

O Se|ect|on Summary pane| vlew/hlde a panel dlsplaylng Lhe evenLs conflgured ln an
AcLlvlLy and Lhe number of samples collecLed per evenL for Lhe lLems you selecL ln a
vlew
O Legend ulsplay a deLalled legend for all sampllng vlews Lach AcLlvlLy resulL evenL
and evenL raLlo ls colorcoded 1he legend explalns whaL each color represenLs
O Lvent summary pane| ulsplay Lhe LoLal number of evenLs collecLed for lLems you
selecL ln a vlew
O Mu|t|processor ulsplay Lhe workload as dlsLrlbuLed across mulLlple processors
SAMLING CVLk 1IML
1 1he Cver Llme vlew dlsplays Lhe samples collecLed for slngle evenL
2 lL enables you Lo ldenLlfy whlch Lhread are runnlng serlally and ln parallel aL any
polnL of Llme
3 Sampllng Cver 1lme vlew can be lnvoked for 1hread9rocess and Module vlews
4 Sampllng over Llme vlew conslsLs of Lwo panels 1he lefL panel dlsplays Lhe names of
Lhe selecLed lLems and Lhe rlghL panel dlsplays Lhe samples collecLed over Llme 1he
rlghL panel ls dlvlded lnLo squares each square represenLlng a unlL of Llme ln
seconds
3 1he color of Lhe squares lndlcaLes Lhe number of samples collecLed for LhaL unlL of
Llme A red square lndlcaLes a large number of samples and a green square lndlcaLes
a small number of samples

I16u8L : Samp1ng 0ver 1me
9age 11 of

1he Cver 1lme vlew can be used Lo gaLher Lhe followlng lnformaLlon
Context Sw|tch|ng Cne can deLermlne lf Lhere ls excesslve conLexL swlLchlng
rocessor Ut|||zat|on Lnables you Lo vlew wheLher processor ls ldle or noL lf syLem
process recelves samples Lhere ls scope for lmprovlng processor uLlllzaLlon aL LhaL
Llme
1empora| |oct|on of hotspots We can see Lhe speclflc perlods of Llme when a large
number of evenLs occurs
1hread Interact|on ?ou can vlew Lhe paLLern of Lhread behavlor and Lhread lnLeracLlon
V|ew|ng the footpr|nt of each thread ?ou can vlew Lhe fooLprlnL of each Lhread on
yper1hreadlng Lechnology enabled processors



















9age 1 of

all Crapbs
IN1kCDUC1ICN
The call graph collector oI the VTune(TM) PerIormance Analyzer collects inIormation about
the program Ilow oI an application, that is, the number oI Iunction calls to some other
Iunction and the amount oI time each Iunction spent executing its code and/or calling other
Iunctions.
A funcLlon can be a
1 CALLL8 A parenL funcLlon LhaL calls Lhe currenL funcLlon
2 CALLLL A chlld funcLlon LhaL ls called by Lhe currenL funcLlon
ln many cases Lhe caller may call Lhe callee from several places (slLes) so call graph also
provldes call lnformaLlon per slLe
LA1UkLS C CALL GkAn
1he followlng are Lhe maln feaLures of Lhe call graph collecLor and vlews
1 Co||ect|on
O Manua| |aunch|ng mode Manually launch your appllcaLlon from Lhe deskLop and
selecL requlred modules of lnLeresL Lo analyze
O DLLLeve| Data Co||ect|on Conflgure Lhe call graph collecLor Lo lnsLrumenL and
analyze flrsLlevel uLLs even when Lhe appllcaLlon lLself cannoL be lnsLrumenLed
O Instrumentat|on f||ter|ng SelecL exacLly whlch funcLlons Lo lnsLrumenL lmprovlng
Lhe speed of Lhe lnsLrumenLed appllcaLlon by uslng lmproved fllLerlng capablllLles
O Mu|t|thread mu|t|process CollecL daLa for more Lhan one process wlLh fully
auLomaLed Lhreadlng and flber supporL
O CCM 1rac|ng 9roflle CCM lnLerface meLhods uslng Lhe call graph collecLor
V|ews
AfLer you collecL call graph daLa uslng Lhe v1une analyzer you can vlew Lhe call graph
proflllng lnformaLlon aL Lhe followlng levels
O GkAn 9rovldes vlsual graphlcal presenLaLlon of Lhe appllcaLlon execuLlon lL
dlsplays Lhe selecLed funcLlon(s) Lhe funcLlons parenLs (callers) lLs chlld funcLlons
(callees) and Llmlng lnformaLlon Lach node {box) ln Lhe graph represenLs a funcLlon
Lach edqe {/ine with on orrow) connecLlng Lwo nodes represenLs Lhe call from Lhe
parenL Lo Lhe chlld funcLlon lor every funcLlon you can Lraverse caller and callee
funcLlons
9age 13 of

1he co// qroph view uses the fo//owinq conventions
O nodes connecLed by Lhlck red edges deslgnaLe funcLlons on Lhe crlLlcal paLh from Lhe
rooL (Lhread)
O 1he Lhlcker Lhe edge Lhe greaLer Lhe Ldge Llme
Dses of this view
O esLlmaLe Lhe performance of your appllcaLlon
O flnd poLenLlal performance boLLlenecks
O Lraverse Lhe crlLlcal paLh whlch ls a paLh wlLh Lhe maxlmum Ldge Llme

I16u8L 2: 6raph vew o1 Ca11 6raph
O CALL LIS1 9rovldes full lnformaLlon on Lhe selecLed or focus funcLlon lLs callers
(parenLs) and callees (chlldren) ln Lhe Lable formaL
1he focus funct|on ls Lhe funcLlon whlch ls currenLly belng vlewed and Lhe focus ls on
LhaL funcLlon lL shows Lhe Lhreads and classes assoclaLed wlLh lL
1he ca||er funct|on ls Lhe funcLlon whlch calls Lhe focused funcLlon Along Lhere are
columns of conLrlbuLlon Ldge Llme Lhreadclass eLc
1he ca||ee funct|on ls Lhe funcLlon whlch ls been called by Lhe focused funcLlon 1here
are also columns almosL same as LhaL of caller funcLlon
9age 1 of


I16u8L 4: Ca11 lst vew o1 Ca11 6raph
O UNC1ICN SUMMAk 9rovldes full lnformaLlon on all Lhe appllcaLlon funcLlons ln
Lhe Lable formaL 1he rows ln Lhe funcLlon summary dlsplay funcLlons wlLh dlfferenL
background colors accordlng Lo Lhe hlerarchlcal poslLlon 1he defaulL vlew shows Lhe
flrsL four Lypes of daLa as follows


I16u8L S: Iuncton Summary vew o1 Ca11 6raph


9age 1 of

3 Accessor|es
lollowlng are Lhe varlous opLlons avallable from Lhe call graph vlew
O ||ter|ng opt|ons Caln dlfferenL perspecLlves on your daLa uslng Lhe wlde range of
fllLerlng opLlons avallable
O unct|on deta|| ConvenlenLly vlew deLalled funcLlon lnformaLlon uslng LoolLlps and
Lhe sLaLus bar
O Un|f|ed Iava support vlew !ava funcLlon calls and Wln32 funcLlon calls ln Lhe same
call graph resulLs
O 1|m|ng opt|ons vlew enrlched Llmlng lnformaLlon wlLh an expanded collecLlon of
walL Llmes for funcLlons and calls 1raverse Self WalL Llme 1oLal WalL Llme Ldge
Llme Ldge WalL Llme and Max paLh from node Lo rooL and from node Lo boLLom
O Node state |nd|cators Ad[usL Lhe color paleLLe for any graph elemenLs and conLrol
node lengLh seLLlngs Lo supporL long funcLlon names node sLaLe lndlcaLors hlghllghL
Lhree dlfferenL Lypes of node sLaLus faclllLaLlng orlenLaLlon wlLhln Lhe graph vlew
O Command access ConLrol a wlde range of opLlons ln Lhe funcLlon summary vlew vla
Lhe funcLlon summary popup menu 1he Loolbar conLalns enhanced feaLures
provldes qulck and easy access Lo Lhe mosL commonly used commands
O Mu|t|p|e undo]redo Make changes Lo Lhe way you vlew daLa Lhen reLurn or
advance forward Lhrough several cycles of changes














9age 1 of

ounter Monitor
IN1kCDUC1ICN
CounLer MonlLor ldenLlfles sysLemlevel lssues ln appllcaLlons lL ls used Lo Lrack sysLem
acLlvlLles when Lhe appllcaLlon runs on Lhe sysLem
CounLer MonlLor collecLs daLa for speclflc performance counLer daLa such as LhaL of an
appllcaLlon an CS or a hardware devlce aL dlfferenL lnLervals of Llme 1he coootet mooltot
collectot monlLors and graphlcally dlsplays Lhe performance counLer daLa
erformance counter ls a feaLure LhaL measure and gaLhers performance relaLed daLa LhaL
represenLs Lhe sLaLe of Lhe sysLem wlLhouL affecLlng Lhe performance of Lhe program
CounLer monlLor also helps you Lo undersLand Lhe causeandeffecL relaLlonshlp beLween
an appllcaLlon and Lhe syLem on whlch Lhe appllcaLlon ls runnlng lf you develop appllcaLlon
speclflc counLers uslng performance dll's Lhe v1une analyzer wlll also monlLor and dlsplay
Lhese counLer values
LA1UkLS C CCUN1Lk MCNI1Ck
1he followlng are Lhe maln feaLures of Lhe counLer monlLor collecLor and vlews
1Co||ect|on
O 1r|gger mechan|sm CreaLe Lrlggers Lo monlLor hardware and sofLware counLers aL
predeLermlned lnLervals accordlng Lo crlLerla LhaL ls seL
A tr|gger ls an evenL LhaL Lells Lhe v1une` 9erformance Analyzer when Lo collecL counLer
daLa 1he v1une analyzer uses Lhe sysLem Llmer as Lhe defaulL Lrlgger lor Lhe sysLem Llmer
performance daLa ls collecLed once per second when Lhe defaulL lnLerval (1000 mllllseconds)
ls used
V|ews
lollowlng are Lhe counLer monlLor vlews Lo help analyze Lhe daLa
kunt|me Data v|ew uurlng runLlme Lhe v1une analyzer generaLes a graph LhaL
shows changes as Lhey happen vlew daLa as you log lL or revlew daLa afLer Lhe run
1hls ls Lhe defaulL vlew whlch runs on compleLlon of an acLlvlLy
Logged Data v|ew ulsplays daLa logged durlng an AcLlvlLy ln Lhe Logged uaLa vlew
daLa from each counLer selecLed for logglng ls charLed wlLh a separaLe llne and color
9age 1 of

Lach llne on Lhe charL represenLs daLa for a speclflc performance counLer 1he peak
lndlcaLes Lhe hlghesL counLer value Movlng Lhe cursor over a counLer on Lhe charL
dlsplays a Lool Llp wlLh Lhe value of Lhe counLer aL LhaL polnL ln Llme durlng daLa
collecLlon

I16u8L 6: logged 0ata vew o1 Counter Montor
1he peaks ln each counLer lndlcaLes Lhe hlghesL counLer acLlvlLy lor example a peak ln Lhe
counLer LhaL measures age au|ts per second lndlcaLes LhaL Lhe mosL page faulLs occurred
aL LhaL polnL ln Llme durlng daLa collecLlon

Legend v|ew Lach llne lncludes a dlsLlncL legend symbol for Lhe correspondlng counLer
represenLlng Lhe polnL aL whlch daLa was Laken 1he verLlcal ? axls represenLs counLer
values (scaled or acLual) whlle Lhe correspondlng Llme ls dlsplayed on Lhe horlzonLal x axls

9age 1 of

Summary Data v|ew ulsplays a sLaLlsLlcal vlew of Lhe counLer daLa
1he Summary uaLa vlew provldes sLaLlsLlcal lnformaLlon for each counLer you selecLed for
dlsplay ln Lhe Logged uaLa vlew 1hls lnformaLlon lncludes
O mlnlmum value
O maxlmum value
O average value
1hls enables you Lo deLermlne whlch values were Lhe mosL acLlve or oLherwlse lnLeresLlng
and drlll down from a Logged uaLa vlew of Lhose values

I16u8L 7: Summary 0ata vew o1 Counter Montor
1he summary daLa for each counLer ls represenLed as a bar dlagram

9age 1 of

where Lhe oppet pott of Lhe dlagram ls Lhe maxlmum value for Lhe counLer (ln Lhe example
1oLal 9rocessor 1lme counLer) Lhe lowet pott ls Lhe mlnlmum value and Lhe mlJJle pott
(vloleL bar ln Lhe example) ls Lhe average counLer value
3 Accessor|es
lollowlng are some opLlons avallable from Lhe counLer monlLor vlew
O Contro| charts Choose a charL sLyle besL sulLed Lo Lhe daLa you wanL Lo vlew uslng
Lhe CharL lx 9roperLles
WCkkING C CCUN1Lk MCNI1Ck
When one selecL an AcLlvlLy wlLh Lhe counLer monlLor collecLor ln Lhe 1unlng 8rowser and
cllck 8un AcLlvlLy Lo begln performance daLa collecLlon Lhe v1une analyzer does Lhe
followlng
1 Launches Lhe speclfled appllcaLlon lf any
2 SLarLs monlLorlng and logglng Lhe counLer values
1he v1une analyzer collecLs performance daLa for all Lhe counLers of a performance
ob[ecL buL dlsplays only Lhe counLers you selecL
3 ulsplays Lhe 8un1lme daLa vlew wlLh a charL showlng Lhe counLer daLa as lL ls belng
collecLed lf Lhe runLlme dlsplay opLlon ls selecLed
4 lf sampllng daLa collecLlon was Lurned on lL also sLarLs collecLlng Llme based or evenL
based sampllng daLa
3 AL Lhe end of an AcLlvlLy run lf counLer monlLor daLa was logged Lhe v1une analyzer
does Lhe followlng
4 CreaLes an AcLlvlLy resulL wlLh Lhe counLer monlLor daLa and shows lL ln Lhe
1unlng 8rowser
4 ulsplays Lhe counLer monlLor Logged uaLa vlew lf Lhe counLer monlLor daLa ls
Lhe only Lype of daLa LhaL was collecLed or prompLs you Lo plck a vlew Lo
open lf mulLlple Lypes of daLa were collecLed








9age 0 of

uning Assistant
IN1kCDUC1ICN
1he lnLel(8) 1unlng AsslsLanL provldes advlce on Lunlng your sysLem resources and
appllcaLlon performance uslng lLs mulLlple knowledge bases Lhe 1unlng AsslsLanL analyzes
Lhe daLa collecLed by Lhe v1une(1M) 9erformance Analyzer ldenLlfles performance lssues
and provldes lnslghLs and Lunlng advlce on Lhe followlng Lypes of daLa
O Sampllng daLa collecLed on supporLed processors
O CounLer monlLor daLa collecLed on supporLed operaLlng sysLems
O C C++ lorLran or !ava* source code
O ulsassembled assembly code
1UNING ASSIS1AN1 CCNCL1S
1he followlng are some key 1unlng AsslsLanL concepLs
O Work|oad All Lhe sofLware LhaL was execuLlng when daLa was collecLed
O Ins|ght An lnslghL ls an observaLlon abouL Lhe performance of your code lL lndlcaLes
a poLenLlal performance problem LhaL could be a boLLleneck Lo your appllcaLlon's
performance
O Adv|ce Advlce ls a posslble soluLlon or recommended workaround (usually a
suggesLlon Lo modlfy Lhe code) Lo remove or avold a performance problem
O ke|evance Score A relevance score ls a heurlsLlc Lo lndlcaLe how relevanL a
parLlcular lnslghL or advlce ls Lo Lhe currenL conLexL lor lnsLance an exLremely hlgh
relevance score for an lnslghL may lndlcaLe a hlgh probablllLy of a performance
boLLleneck
1he 1unlng AsslsLanL provldes Lunlng advlce for code processes/modules/funcLlons or Llme
ranges LhaL you selecL ln source sampllng or counLer monlLor vlews lf you provlde symbol
lnformaLlon Lhe 1unlng AsslsLanL wlndow provldes llnks from your funcLlon names dlrecLly
Lo Lhe correspondlng code secLlon ln Source vlew
LA1UkLS C 1UNING ASSIS1AN1
1he lnLel(8) 1unlng AsslsLanL has Lhe followlng feaLures Lo enable analyzlng Lhe performance
of your appllcaLlon
O 9rovldes lnslghLs and advlce on poLenLlal performance problems by analyzlng
sampllng daLa collecLed on supporLed processors (See Lhe 8elease noLes for a
compleLe llsL of processors for whlch Lhe 1unlng AsslsLanL can provlde lnslghLs and
advlce) ?ou can use Lhe lnslghLs and advlce Lo make algorlLhmlc changes Lo your
appllcaLlon so Lhe processor can execuLe your appllcaLlon more efflclenLly
9age 1 of

O ConLalns knowledge bases Lo supporL yper1hreadlng 1echnology
O Lnables you Lo compare Lwo or Lhree AcLlvlLy resulLs
O 9rovldes llnks from funcLlon names dlrecLly Lo Lhe correspondlng code secLlon ln
source vlew when you provlde symbol lnformaLlon
O 9rovldes advlce on performance counLer daLa and dlsassembly code
O 9rovldes sLaLlc assembly advlce
O Culdes you Lhrough Lhe key sLeps of performance Lunlng meLhodology
O 9rovldes Lhe ablllLy Lo exporL Lhe Lunlng advlce reporL Lo a csv (comma separaLed
values) LexL flle for vlewlng and edlLlng uslng a dlfferenL appllcaLlon such as
MlcrosofL* Lxcel
UNDLkS1ANDING 1UNING ML1nCDCLCG
1 SystemLeve| 1un|ng 1he maln ob[ecLlve of sysLemlevel Lunlng ls Lo opLlmlze Lhe
uLlllzaLlon of sysLem resources 1he Lunlng speeds up appllcaLlon performance by
lmprovlng Lhe way Lhe appllcaLlon lnLeracLs wlLh Lhe syLem 1hls Lunlng ls effecLlve
for l/C appllcaLlons
2 App||cat|onLeve| 1un|ng 1he maln purpose of appllcaLlonlevel Lunlng ls Lo reduce
Lhe execuLlon Llme of an appllcaLlon 1hls can be achleved by lmprovlng Lhe
algorlLhms of Lhe appllcaLlons lmplemenLlng Lhreads and by uslng A9ls
3 M|croarch|tectureLeve| 1un|ng lncreases Lhe performance of appllcaLlon by
lmprovlng Lhe way an appllcaLlon runs on a processor 1hls Lype of Lunlng ls used
wlLh processorlnLenslve appllcaLlons
S1kA1LGILS Ck IMkCVING LkCkMANCL C ALICA1ICN
8a|anc|ng InputCutput Lnables Lo speed up appllcaLlon when processor uLlllzaLlon
ls low 9rocessor uLlllzaLlon drops when Lhe processor ls walLlng for l/C Lo compleLe
need Lo make changes ln app durlng sysLem level and appllcaLlonlevel Lunlng
Improv|ng thread|ng mode| 8y addlng mulLlLhreadlng Lo slngleLhreaded app
lmprove efflclency of app by lncreaslng processor uLlllzaLlon
Improv|ng the eff|c|ency of computat|on Speed up appllcaLlon by maklng changes
Lo Lhe appllcaLlon Lo accompllsh Lhe same amounL of work by uslng less
compuLaLlon
1LS C ADVICL
Samp||ng based adv|ce 1unlng asslsLanL auLomaLlcally analyzes Lhe sampllng
daLaldenLlfles performance lssues and provldes lnslghLs on Lhe lssues When one
cllck an lnslghL Lhe ,ore lnformotion wlndow provldes addlLlonal lnformaLlon 1hls
wlndow conLalns ke/evonce sco/e LhaL can be use Lo vlew Lhe relevance of a
parLlcular lnslghL Lo performance lssues
9age of


I16u8L 8: ^dvce Wndow {Showng Samp1ng-based advce) o1 1unng
^ssstant ^dvce
Counter Mon|torbased adv|ce 1unlng asslsLanL performs counLer analysls based
on all counLers measured ln acLlvlLy AfLer analysls 1A dlsplays lnslghLs lnLo poLenLlal
performance boLLlenecks
Sourcebased adv|ce 1A uses a compller Lechnology for sourcebased advlce whlch
enables you Lo speed up Lhe execuLlon of code 8uL lL ls llmlLed Lo CC++ and !ava
appllcaLlons
Stat|c Assemb|e ena|t|es v1une analyze code aL assembly language level 1he Lwo
caLegorles of lnformaLlon LhaL 1A dlsplays are
1 9enalLy lndlcaLes a speclflc problem and Lhe effecL of Lhe problem on
performance of code
2 Warnlng lndlcaLes poLenLlal problems LhaL mlghL degrade Lhe performance
INCkMA1ICN 1nA1 1UNING ASSIS1AN1 kCVIDLS INCLUDLS
INSIGn1S lndlcaLes Lhe problem LhaL could be hlnderlng Lhe performance of Lhe
appllcaLlon varlous caLegorles of lnslghLs are
1 1op lnslghLs 1haL are esLlmaLed Lo have slgnlflcanL lmpacLs on performance
Lnables Lo ldenLlfy Lhe maxlmum opLlmlzaLlon LhaL one can achleve for Lhe
appllcaLlon
Workload lnslghLs Are performance lssues for all modules and
processes(See flg 8)
3 Module lnslghLs locus on performance lssues for Lhe modules ln an
appllcaLlon(See flg 8)
9age 3 of

oLspoLs lnslghLs lnslghLs on performance lssues based on funcLlons LhaL are
sorLed by percenLage of C9u Llme
SysLem lnfo Summarlzes Lhe feaLures LhaL Lhe sysLem uses such as sped of
processor and Lhe name of Lhe operaLlng sysLem
SLaLlc Analysls vlew lnformaLlon abouL posslble opLlmlzaLlons Lo lmprove
app performance

I16u8L : More 1n1ormation Wndow o1 1unng ^sstant ^dvce
kLLLVANCL SCALL lndlcaLes Lhe relevance of Lhe lnslghL or advlce Lo a parLlcular
performance lssue lor example a hlgh relevance score lndlcaLes LhaL Lhe effecL of
Lhe problem on Lhe appllcaLlon ls slgnlflcanL or 100(See flg 9)
1UNING ASSIS1AN1 ADVICL 9osslble soluLlon Lo remove or avold a problem Cne
can cllck on llnks as shown ln flg 8 Lo geL advlce






9age of

eferences
1he followlng are Lhe references whlch have been used for documenLaLlon purpose
ne|p f||e lnLel v1une sofLware help flle ls used
8ooks
1 lnLel v1une 9erformance Analyzer LssenLlals (AuLhor !ames 8elnders)
2 3
rd
SemesLer lnLel v1une (8y nll1)
Webs|tes
1 wwwlnLelcom
2 wwwhlperlsmcom

Das könnte Ihnen auch gefallen