6/5/2017 IndexinginDBMSOrderedIndicesPrimaryIndexDenseIndexSparseIndexSecondaryIndexMultilevelIndicesClusteringIndexinDatabase

IntroductionofIndexing

Themaingoalofdesigningthedatabaseisfasteraccesstoanydatainthedatabaseandquicker
insert/delete/updatetoanydata.Thisisbecausenoonelikeswaiting.Whenadatabaseisveryhuge,evena
smallesttransactionwilltaketimetoperformtheaction.Inordertoreducethetimespentintransactions,Indexes
areused.Indexesaresimilartobookcataloguesinlibraryorevenlikeanindexinabook.Whatitdoes?Itmakes
oursearchsimplerandquicker.SameconceptisappliedhereinDBMStoaccessthefilesfromthememory.

WhenrecordsarestoredintheprimarymemorylikeRAM,accessingthemisveryeasyandquick.Butrecordsare
notlimitedinnumberstostoreinRAM.Theyareveryhugeandwehavetostoreitinthesecondarymemorieslike
harddisk.Aswehaveseenalready,inmemorywecannotstorerecordslikeweseetables.Theyarestoredin
theformoffilesindifferentdatablocks.Eachblockiscapableofstoringoneormorerecordsdependingonits
size.

Whenwehavetoretrieveanyrequireddataorperformsometransactiononthosedata,wehavetopullthem
frommemory,performthetransactionandsavethembacktothememory.Inordertodoalltheseactivities,we
needtohavealinkbetweentherecordsandthedatablockssothatwecanknowwheretheserecordsarestored.
Thislinkbetweentherecordsandthedatablockiscalledindex.Itactslikeabridgebetweentherecordsandthe
datablock.

Howdowecreatetheseindexes?Howtheseindexeshelptoaccessthedata?Islinkingbetweenrecordsand
datablockaddressenoughtogivebetterperformance?Answersallthesequestionsarelearntinthisarticle.

Howdoweindexinabook?Welistmaintopicsfirstandunderthatwegroupdifferentsubtopicright?Wedothe
samethinginthedatabasetoo.Eachtablewillhaveuniquecolumnorprimarykeycolumnwhichuniquely
determineseachrecordinthetable.Mostofthetime,weusethisprimarykeytocreateindex.Sometimes,wewill
havetofetchtherecordsbasedonothercolumnsinthetablewhicharenotprimarykey.Insuchcaseswecreate
indexonthosecolumns.Butwhatisthisindex?Indexindatabasesisthepointertotheblockaddressinthe
memory.Butthesepointersarestoredas(column,block_address)format.

OrderedIndices

Imaginewehaveastudenttablewiththousandsofrecords,eachofwhichis10byteslong.ImaginetheirIDsstart
from12,3andgoeson.AndwehavetosearchstudentwithID678.Inanormaldatabasewithnoindex,it
searchesthediskblockfromthebeginningtillitreaches678.SotheDBMSwillreachthisrecordafterreading
677*10=6770bytes.ButifwehaveindexonIDcolumn,thentheaddressofthelocationwillbestoredaseach
recordas(1,200),(2,201)(678,879)andsoon.Onecanimagineitasasmallertablewithindexcolumnand
addresscolumn.NowifwewanttosearchrecordwithID678,thenitwillsearchusingindexes.i.e.hereitwill
traverseonly677*2=1354byteswhichverylesscomparedtoearlierone.Henceretrievingtherecordfromthe
diskbecomesfaster.Mostofthecasestheseindexesaresortedandkepttomakesearchingfaster.Iftheindexes
aresorted,thenitiscalledasorderedindices.

PrimaryIndex

https://www.tutorialcup.com/dbms/indexing.htm 1/7
6/5/2017 IndexinginDBMSOrderedIndicesPrimaryIndexDenseIndexSparseIndexSecondaryIndexMultilevelIndicesClusteringIndexinDatabase

IftheindexiscreatedontheprimarykeyofthetablethenitiscalledasPrimaryIndexing.Sincetheseprimary
keysareuniquetoeachrecordandithas1:1relationbetweentherecords,itismucheasiertofetchtherecord
usingit.Also,theseprimarykeyarekeptinsortedformwhichhelpsinperformanceofthetransactions.The
primaryindexingisoftwotypesDenseIndexandSparseIndex.

1.DenseIndex

Inthiscase,indexingiscreatedforprimarykeyaswellasonthecolumnsonwhichweperformtransactions.That
means,usercanfirequerynotonlybasedonprimarykeycolumn.Hecanquerybasedonanycolumnsinthe
tableaccordingtohisrequirement.Butcreatingindexonlyonprimarykeywillnothelpinthiscase.Henceindex
onallthesearchkeycolumnsarestored.Thismethodiscalleddenseindex.

Forexample,StudentcanbesearchedbasedonhisIDwhichisaprimarykey.Inaddition,wesearchforstudent
byhisfirstname,lastname,particularagegroup,residinginsomeplace,optedforsomecourseetc.Thatmeans
mostofthecolumnsinthetablecanbeusedforsearchingthestudentbasedondifferentcriteria.Butifwehave
indexonhisID,othersearcheswillnotbeefficient.Henceindexonothersearchcolumnsarealsostoredtomake
thefetchfaster.

Thoughitaddressesquicksearchonanysearchkey,thespaceusedforindexandaddressbecomesoverheadin
thememory.Herethe(index,address)becomesalmostsameas(tablerecords,address).Hencemorespaceis
consumedtostoretheindexesastherecordsizeincreases.

https://www.tutorialcup.com/dbms/indexing.htm 2/7
6/5/2017 IndexinginDBMSOrderedIndicesPrimaryIndexDenseIndexSparseIndexSecondaryIndexMultilevelIndicesClusteringIndexinDatabase

2.SparseIndex

Inordertoaddresstheissuesofdenseindexing,sparseindexingisintroduced.Inthismethodofindexing,range
ofindexcolumnsstorethesamedatablockaddress.Andwhendataistoberetrieved,theblockaddresswillbe
fetchedlinearlytillwegettherequesteddata.

Letusseehowaboveexampleofdenseindexisconvertedintosparseindex.

Inabovediagramwecansee,wehavenotstoredtheindexesforalltherecords,insteadonlyfor3records
indexesarestored.NowifwehavetosearchastudentwithID102,thentheaddressfortheIDlessthanorequal
to102issearchedwhichreturnstheaddressofID100.Thisaddresslocationisthenfetchedlinearlytillweget
therecordsfor102.Henceitmakesthesearchingfasterandalsoreducesthestoragespaceforindexes.

Therangeofcolumnvaluestostoretheindexaddressescanbeincreasedordecreaseddependingonthe
numberofrecordinthetable.Themaingoalofthismethodshouldbemoreefficientsearchwithlessmemory
space.

Butifwehaveveryhugetable,thenifweprovideverylargerangebetweenthecolumnswillnotwork.Wewill
havetodividethecolumnrangesconsiderablyshorter.Inthissituation,(index,address)mappingfilesizegrows
likewehaveseeninthedenseindexing.

SecondaryIndex

Inthesparseindexing,asthetablesizegrows,the(index,address)mappingfilesizealsogrows.Inthememory,
usuallythesemappingsarekeptintheprimarymemorysothataddressfetchshouldbefaster.Andlatterthe
actualdataissearchedfromthesecondarymemorybasedontheaddressgotfrommapping.Ifthesizeofthis
mappinggrows,fetchingtheaddressitselfbecomesslower.Hencesparseindexwillnotbeefficient.Inorderto
overcomethisproblemnextversionofsparseindexingisintroducedi.e.SecondaryIndexing.

Inthismethod,anotherlevelofindexingisintroducedtoreducethe(index,address)mappingsize.Thatmeans
initiallyhugerangeforthecolumnsareselectedsothatfirstlevelofmappingsizeissmall.Theneachrangeis

https://www.tutorialcup.com/dbms/indexing.htm 3/7
6/5/2017 IndexinginDBMSOrderedIndicesPrimaryIndexDenseIndexSparseIndexSecondaryIndexMultilevelIndicesClusteringIndexinDatabase

furtherdividedintosmallerranges.Firstlevelofmappingisstoredintheprimarymemorysothataddressfetchis
faster.Secondarylevelofmappingandtheactualdataarestoredinthesecondarymemoryharddisk.

Intheabovediagram,wecanseethatcolumnsaredividedintogroupsof100sfirst.Thesegroupsarestoredin
theprimarymemory.Inthesecondarymemory,thesegroupsarefurtherdividedintosubgroups.Actualdata
recordsarethenstoredinthephysicalmemory.Wecannoticethat,addressindexinthefirstlevelispointingto
thefirstaddressinthesecondarylevelandeachsecondaryindexaddressesarepointingtothefirstaddressin
thedatablock.Ifwehavetosearchanydatainbetweenthesevalues,thenitwillsearchthecorresponding
addressfromfirstandsecondlevelrespectively.Thenitwillgototheaddressinthedatablocksandperform
linearsearchtogetthedata.

Forexample,ifithastosearch111intheabovediagramexample,itwillsearchthemax(111)<=111inthefirst
levelindex.Itwillget100atthislevel.Theninthesecondaryindexlevel,againitdoesmax(111)<=111,andgets
110.Nowitgoestodatablockwithaddress110andstartssearchingeachrecordtillitgets111.Thisishowa
searchisdoneinthismethod.Inserting/deleting/updatingisalsodoneinsamemanner.

MultilevelIndexing

Inthismethod,wecanseethatindexmappinggrowthisreducedtoconsiderableamount.Butthismethodcan
alsohavesameproblemasthetablesizeincreases.Inordertoovercomethis,wecanintroducemultiplelevels
betweenprimarymemoryandsecondarymemory.Thismethodisalsoknownasmultilevelindexing.Inthis
methodnumberofsecondarylevelindexistwoormore.

ClusteringIndex

Insomecases,theindexiscreatedonnonprimarykeycolumnswhichmaynotbeuniqueforeachrecord.In
suchcases,inordertoidentifytherecordsfaster,wewillgrouptwoormorecolumnstogethertogettheunique

https://www.tutorialcup.com/dbms/indexing.htm 4/7
6/5/2017 IndexinginDBMSOrderedIndicesPrimaryIndexDenseIndexSparseIndexSecondaryIndexMultilevelIndicesClusteringIndexinDatabase

valuesandcreateindexoutofthem.Thismethodisknownasclusteringindex.Basically,recordswithsimilar
characteristicsaregroupedtogetherandindexesarecreatedforthesegroups.

Forexample,studentsstudyingineachsemesteraregroupedtogether.i.e.1stSemesterstudents,2ndsemester
students,3rdsemesterstudentsetcaregrouped.

Inabovediagramwecanseethat,indexesarecreatedforeachsemesterintheindexfile.Inthedatablock,the
studentsofeachsemesteraregroupedtogethertoformthecluster.Theaddressintheindexfilepointstothe
beginningofeachcluster.Inthedatablocks,requestedstudentIDisthensearchinsequentially.

Newrecordsareinsertedintotheclustersbasedontheirgroup.Inabovecase,ifanewstudentjoins
3rdsemester,thenhisrecordisinsertedintothesemester3clusterinthesecondarymemory.Sameisdonewith
updateanddelete.

Ifthereisshortofmemoryinanycluster,newdatablocksareaddedtothatcluster.

https://www.tutorialcup.com/dbms/indexing.htm 5/7
6/5/2017 IndexinginDBMSOrderedIndicesPrimaryIndexDenseIndexSparseIndexSecondaryIndexMultilevelIndicesClusteringIndexinDatabase

Thismethodoffileorganizationisbettercomparedtoothermethodsasitprovidescleandistributionofrecords,
andhencemakingsearcheasierandfaster.Butineachcluster,therewouldbeunusedspaceleft.Henceitwill
takemorememorycomparedtoothermethods.

Summary

https://www.tutorialcup.com/dbms/indexing.htm 6/7
6/5/2017 IndexinginDBMSOrderedIndicesPrimaryIndexDenseIndexSparseIndexSecondaryIndexMultilevelIndicesClusteringIndexinDatabase

https://www.tutorialcup.com/dbms/indexing.htm 7/7