CSE414FinalExamination

Name:__________Solution____________

Topic Points Max

I RelationalData 30

II DBApplications 36

III SemistructuredData 12

IV DBMSImplementation 48

V BigData 26

Total 152

• Theexamisclosedbook.• Noelectronicdevicesmaybeused.• Youareallowedtwoletter-sizedpagesofnotes(bothsides,8+ptfont).

• Allquestionshaveoneanswerunlessstatedotherwise.• Questionswithoneanswerareworth2points.• Questionswithmanyanswersareworth1pointperavailableoption(4-6pointstotal).

• Youhave1hourand50minutestocompletethetest.• Ifyouarenotdoneafter1hourand30minutes,pleasestayinyourseatuntiltheend.

I.RelationalData

Considerthefollowingschemaforasimplesocialnetwork.Itmaintainsacollectionof

users,identifiedbyunique“handles”(shortnames).

User(handle, name, home_city, bio) Friend(handle1, handle2) Message(handle, text, from_city)

Thenextfewproblemswillconsiderqueriesforansweringthefollowingquestions:

A. Findthehandlesofuserswhowrotemessageswhileincitiesotherthantheoneinwhichtheylive.

B. Findthehandlesofuserswhoarefriendswithsomeonewhoisfriendswiththem.

C. FindthenamesofuserswhoeitherliveinKansasCityorhavewrittenamessagefromKansasCity.

D. FindthenamesofuserswhobothliveinKansasCityandhavewrittenamessagefromKansasCity.

Afteryouhavereadthesequestions,turnthepage….

(Therearenoproblemsonthispage.)

Foreachextendedrelationalalgebraexpressionbelow,indicatewhichofthequestions

above(A–D),ifany,itanswers.Ifnoquestionmatches,thenleavethespaceblank.

1. δ(πhandle1σhandle1=handle3[Friend⋈(ρhandle2,handle3Friend)]) ________B_______

(Recallthattheρoperatorsimplyrenamesthecolumnsofatable.)

2. δ(πhandle[δ(πhandle,from_cityMessage)–πhandle,home_cityUser]) ________A_______

3. δ(πname[([πhandleσhome_city=‘KansasCity’User]∪ ________C_______

[πhandleσfrom_city=‘KansasCity’Message])⋈User])

4. δ[πhandleσfrom_city<>home_city(User⋈Message)] ________A_______

ForeachSQLquerybelow,indicatewhichofthequestionsfrombefore(A–D),ifany,it

answers.Ifnoquestionmatches,thenleavethespaceblank.

5. SELECT name ________C_______ FROM User U WHERE home_city = ‘Kansas City’ OR EXISTS (SELECT name FROM Message WHERE handle = U.handle AND from_city = ‘Kansas City’)

6. (SELECT name FROM User ________D_______ WHERE home_city = ‘Kansas City’) INTERSECT (SELECT name FROM User U, Message M WHERE U.handle = M.handle AND from_city = ‘Kansas City’)

7. SELECT DISTINCT U.handle _______A________ FROM User U, Message M WHERE U.handle = Message.handle AND from_city <> home_city

8. SELECT F1.handle1 ________________ FROM Friend F1, Friend F2 WHERE F1.handle2 = F2.handle2 AND F1.handle1 = F2.handle1 GROUP BY F1.handle1

ForeachDatalogquerybelow,indicatewhichofthequestionsfromearlier(A–D),ifany,it

answers.Ifnoquestionmatches,thenleavethespaceblank.

Ineachitem,theruleoftheformQN,whereNisanumber,definesthequeryinquestion.

9. Q1(h) :- User(h, _, hc, _), ________A_______ Message(h, _, fc), hc <> fc.

10. Q2(h) :- User(h, _, “Kansas City”, _). ________________ Q2(h) :- User(h, _, hc, _), Message(h, _, fc), hc = “Kansas City”.

11. Q3(h) :- User(h, _, “Kansas City”, _), ________D_______ Message(h, _, “Kansas City”).

12. FoF(h1, h2, 1) :- Friend(h1, h2). ________B_______ FoF(h1, h2, d) := FoF(h1, h3, d-1), Friend(h3, h2). Q4(h) :- FoF(h, h, 2).

13. ConsiderthefollowingSQLquery:

SELECT handle, bio, min(home_city) FROM User NATURAL JOIN Message GROUP BY handle, from_city

Whichsectionofthequeryisillegal?

handle bio

min(home_city) from_city

14. Considerthefollowingexpressioninstandardrelationalalgebra:

πfrom_city[Message⋈(σhandle>‘c’User–σhandle>‘d’User)]

Canthesameresultbecomputedusingasimpleselect-from-wherequery(i.e.,onewith

nosubqueries,grouping,oraggregation)?

Yes No

15. Whichofthefollowingisanadvantageofusinganauto-generated,uniqueintegerIDastheprimarykeyforUserinsteadofthehandle?(Choosethebestanswer.)

usesmuchlessdiskspace makessearchesbyprimarykey

muchmoreefficient

makeshandlechanges makeshashindexesanoption

muchmoreefficient insteadofonlyB+treeindexes

II.DBApplications

1. Whichofthefollowingaretupleconstraintsthatcanbeautomaticallycheckedbya

DBMSwhentuplesarechangedoraddedtothetable?Circleallthatapply.

valueincolumnisastring valueincolumnisthekeyofatupleintableT

valueincolumnisnotnull valueincolumnisapositiveinteger

valueincolumnisunique valuesintwocolumnsaredifferent

inthatcolumnofthetable

2. SupposethatR(a,b,c)andS(c,d)arerelations.IfRhasadependencyb→c,thenthesamedependencymustexistin:

R⨉S: Yes No

R⋈S: Yes No

3. ForeachJDBCstatementtypebelow,towhichmethoddoyoupassthestringdescribing

yourSQLquery?(RecallthatStatementscanexecutemultiplequeries.)

Statement: createStatement executeQuery PreparedStatement: prepareStatement executeQuery

4. UsingSQLServer,ifoneJDBCConnectionsetsitsisolationleveltoreaduncommitted,onlythatconnectionmayseebehaviorthatisnotserializable?

Yes No

5. Whichtierofa3-tieredarchitectureismostlikelytobecometheperformance

bottleneckasyouscaletoaverylargenumberofusers?

browsers harddisks

webservers DBserver

ConsiderthefollowingSQLschema:

CREATE TABLE Company( name VARCHAR(100) PRIMARY KEY, address VARCHAR(200)); CREATE TABLE Product( name VARCHAR(100) PRIMARY KEY, price FLOAT, made_by VARCHAR(100) FOREIGN KEY REFERENCES Company);

6. WhichofthefollowingE/Rdiagramscouldproducethatschema?

7. WhichofthefollowingdescribestherelationshipbetweenProductsandCompanies?

onetoone onetomany

manytoone manytomany

8. WhichofthefollowingtypesofrelationshipsbetweenProductsandCompaniescould

berepresentedinSQLwithoutanadditionaltable?Circleallthatapply.

onetoone onetomany

manytoone manytomany

ThenexttwoquestionsrefertotheschemaforUserandMessagethatwesawinPartI.

9. Whichofthefollowingchangestothedataforasingleuserwould,mostlikely,require

changingmorerowsofthedatabase?

changinghandle changingname

10. Whichofthefollowingwouldbethebestwaytoreducethenumberofrowsthatneedtobechangedthecaseabove?

addanindexonMessage(handle) use(handle, name)astheprimarykeyofUser

addanindexonMessage(text) useanauto-generated,uniqueID

astheprimarykeyofUser

Supposethatwechangedtheschemafromearliersothat(1)theauthor’snameisincluded

(inadditiontotheirhandle)and(2)thecity’szipcodeisincluded:

Message(handle, name, text, from_city, city_zip)

11. Supposethatwealloweduserstochangeboththeirnamesandhandles.Eachitembelowgivesonechoiceofthemanynamesandhandlesauserhashadthatwecould

storewiththemessage.Inwhichsituations,wouldhandle→nameinMessagestillbeafunctionaldependency?Circleallthatapply.

currenthandle currenthandle

nameattimewritten currentname

handleattimewritten firsthandleofauthor

nameattimewritten currentname

12. WhichofthefollowingcouldresultfromaBCNFdecompositionofthistable?Circleallthatapply.

City(name, zip) Message(name, text, from_city) Author(handle, name) Message(handle, text, from_city)

13. Whichofthesewouldmakesensetoaddtotheschema?Circleallthatapply.

City Author

III.SemistructuredData

SupposethatwestoreallthedataforoursocialnetworkinasingledatasetofUsers:

[{“handle”: “biebs”, “name”: “Justin Bieber”, “home_city”: “Somewhere, Canada”, “bio”: “…”, “friends”: [“kimkardashian”, “shaq”, …], “messages”: [ {“text”: “:-* :-* :-* :-*”, “from_city”: “Los Angeles, CA”}, {“text”: “New. Music. Friday.”, “from_city”: “Los Angeles, CA”}, … ]} … ]

ConsiderthefollowingSQL++query:

SELECT DISTINCT m.from_city, n.from_city FROM Users u, u.messages m, Users v, v.messages n WHERE u.handle <> v.handle AND m.text = n.text

1. WhichoftheSQLqueriesbelowreturnthesameresultastheSQL++queryabove?

(EachSQLqueryusesthesameschemaforUserandMessageaswesawinPartI.)Circleallthatapply.

SELECT DISTINCT m.from_city, n.from_city FROM User u, Message m, User v, Message n WHERE u.handle = m.handle AND v.handle = m.handle AND u.handle <> v.handle AND m.text = n.text SELECT DISTINCT m.from_city, n.from_city FROM Message m, Message n WHERE m.handle <> n.handle AND m.text <> n.text SELECT m.from_city, n.from_city FROM Message m, Message n WHERE m.handle <> n.handle AND m.text <> n.text GROUP BY m.from_city, n.from_city SELECT DISTINCT m.from_city, n.from_city FROM Message m, Message n, User v WHERE m.handle <> v.handle AND m.text = n.text

ConsiderthefollowingSQL++query:

SELECT max(coll_count(m)) FROM User u, u.messages m WHERE m.text > 3 AND u.handle = “biebs” ORDER BY u.name

2. Whichpartsofthequeryaboveareillegal?Circleallthatapply.

coll_count(m) ORDER BY u.name

u.handle = “biebs” m.text > 3

ConsiderthefollowingSQL++query:

SELECT u.handle, count(*) FROM User u, u.messages m WHERE u.home_city = ‘Los Angeles, CA’ GROUP BY u.handle

3. WhichofthefollowingSQL++queriesreturnthesameresult?Circleallthatapply.

SELECT u.handle, coll_count(u.messages) FROM User u WHERE u.home_city = ‘Los Angeles, CA’ SELECT u.handle, coll_count((SELECT * FROM u.messages)) FROM User u WHERE u.home_city = ‘Los Angeles, CA’

4. WhichofthefollowingSQLqueriesreturnthesameresult?Circleallthatapply.

SELECT u.handle, count(*) FROM User u, Message m WHERE u.handle = m.handle AND u.home_city = ‘Los Angeles, CA’ GROUP BY u.handle

SELECT u.handle, count(*) FROM User u NATURAL JOIN Message WHERE u.home_city = ‘Los Angeles, CA' GROUP BY u.handle

IV.DBMSImplementation

1. SupposethatA=100,transactionT1wantstoincreaseAby100,andT2wantsto

doubleA.ConsiderthescheduleR1(A),R2(A),W1(A),W2(A).Whichofthefollowing

propertiesholdforthisschedule?Circleallthatapply.

transactionsareisolated serializable

followsstrict2PLlocking conflictserializable

2. SupposethatT1andT2areaccessingelementBinaglobally-distributeddatabase

fromoppositesidesoftheplanet.Whichofthefollowingismostdifficulttosupport?

R1(B)beforeR2(B) R1(B)beforeW2(B)

W1(B)beforeR2(B) W1(B)beforeW2(B)

3. Intheabsenceofrollbacks,whichofthefollowingensuresthatlockingresultina

serializableschedule?Circleallthatapply.

two-phaselocking usingreadandwritelocks

stricttwo-phaselocking usingread,pending,reserved,and

exclusivelocks

4. WhichoftheACIDpropertiesaremostatriskofbeingviolatediftheDBMScrashesmid-

waythroughcommittingresultstodisk.Circletwo.

atomicity consistency

isolation durability

5. Intheabsenceofrollbacks,whichofthefollowingcanbeusedtodeterminewhethera

scheduleisconflictserializable?Circleallthatapply.

checkforcyclesinprecedencegraph

simulatea2PLschemeandcheckfordeadlocks

simulatestrict2PLandcheckfordeadlocks

seeifyoucantochangethescheduleintoaserialorder

byswappingadjacentelements

6. Ifaddingmoreclientsisdecreasingtherateoftransactionspersecond,whichofthe

followingtechniqueswilllikelyincreasetheratewiththeleastpossibilityofbugs?

readcommittedisolationlevel addmorewebservers

readuncommittedisolationlevel admissioncontrol

7. SupposewearecomputingR⋈S,withbothtableslarge(1M+rows).Theestimatedcostofablocknestedloopjoinisapproximatelycutinhalfif(circleallthatapply):

numberoftuplesofRhalved tuplesperblockofRisdoubled

withnumberofblocksunchanged withnumberoftuplesunchanged

numberoftuplesofSishalved tuplesperblockofSisdoubled

withnumberofblocksunchanged withnumberoftuplesunchanged

8. Comparedtoregular(tuple-based)nestedloopjoin,theestimatedcostofablocknested

loopjoinis:

alwaysfaster neverfaster

sometimesfaster,sometimesslower neverslower

9. Ifeachblockcontainsonlyasingletuple,thencomparedtoablocknestedloopjoin,the

estimatedcostofanindexedjoinis:

alwaysfaster neverfaster

sometimesfaster,sometimesslower neverslower

10. Supposethattheprimarykeyoftableisanauto-generated,UUIDwithseeminglyrandomvalues.Comparedtoanunclusteredindexontheprimarykey,theclustered

indexcreatedautomaticallybySQLservermakeseachquery:

alwaysfaster alwaysslower

sometimesfaster,sometimesslower alwaysthesame

Inthenextfewproblems,wewillcomputethecost

ofthequeryplanontheright,whichfindstheresult

oftheexpressionπe,h(R⋈S⋈T)overthetablesR(e,f),S(f,g),andT(g,h).(Bothjoinsnatural.)

Theplanontherightisstillincomplete,however,becausethechoiceofalgorithmfortwoparts,JoinA

andJoinB,havenotyetbeenfilledin.

Inthenextfewproblems(startingonthenextpage),

wewillconsiderthecostofthisplanwithdifferent

choicesofalgorithmsforthesetwopoints.

Youcanassumethefollowingstatisticsaboutthesetables:

Table #tuples #blocks

R 1,000 100

S 5,000 200

T 100,000 10,000

R⋈S 5,000 20

Furthermore,youcanassumethefollowingstatisticsabouttheircolumns:

Column #distinct Low High

R.f 100 1 1000

S.f 1000 1 2000

S.g 5000 1 2,000

T.g 1000 1 10,000

UsefulFormulas

EstimatedcostofXjoinY:

• Usingablocknestedloop,thecostisB(X)+B(X)B(Y),whereB(X)is#blocksinX.• UsingaclusteredindexonY(A),thecostisB(X)+T(X)B(Y)*E,whereT(X)is#

tuplesinX,andEistheselectivityoftheconditionA = c.• UsinganunclusteredindexonY(A),thecostisB(X)+T(X)T(Y)*E.

Estimatedselectivityofconditions:

• ForA = c,theselectivityis1/(#distinctvaluesofA)• ForA < c,theselectivityis(c-lowestvalueofA)/(highest-lowestvalueofA)

11. WhatistheestimatedcostofJoinBintheplanifweimplementitwithablocknestedloopjoin?

20+20*10,000=200,020

12. WhatistheestimatedcostofJoinBtheifweuseanindexedjoin?AssumethatwehaveaclusteredindexonT(g).

20+5,000*10,000/1000=50,020

13. WhatistheestimatedcostofJoinAintheplanifweimplementitwithablocknestedloopjoin?

100+100*200=20,100

14. WhatistheestimatedcostofJoinAifweuseanindexedjoin?AssumethatwehaveanunclusteredindexonS(f).

100+1,000*5,000/1000=5,100

15. WhatisthetotalcostofthisplanifweusethebestchoiceofjoinalgorithmforAandB?

Inadditiontothetwojoinsabove,weneedtowriteR⋈Stodisk,whichcosts20IOs.Thefinalprojectionhasnocost.Thus,thetotalcostis55,170.

YouareimplementingaDBapplicationusingSQLServer.Atfirst,asyouaddmoreusers,

youseeaniceincreaseincompletedtransactionspersecond.However,afterreachinga

highlevelofsimultaneoususers,youfindthatcompletedtransactionspersecondstarts

droppingasyouaddmoreusers.

YoudeterminethattheproblemisSQLServer’srow-levellocking.Toimprove

performance,youwouldliketorunsomequeriesusingtable-levellockinginstead.

Yourideaisasfollows.Youcreateaspecialtable:

Tables(name VARCHAR(100), last_write_by INT);

Itwillcontainonerowforeachtable,indicatingthenameofthattableandthenumberof

thetransactionthatlastwrotetoit.Eachtransactionwillstartbyreadingtherowsof

Tablesforthetablesthatitswantstoreadandwritingtherowsforthetablesthatitwantstowrite,whichwillrequireSqlServertoacquirereadorwritelocksonthoserows.Inthat

manner,therow-levellocksinSQLServerTablesforactastable-levellocks.

16. Afterthetransactionhasacquiredtheappropriatetablelocksanditexecutesitsoperationsagainsttheactualtables,willSQLServerstillacquirelocksontherowsof

thosetables?

Yes No

17. Toimproveperformance,youdecidetouseasecondDBconnectiontoexecutethequeryonceyouhaveacquiredthetablelocksusingyourprimaryDBconnection.What

istheweakestisolationlevelthatyoucansafelyuseforthissecondconnectionifyou

stillwantyourtransactionstobeACID?

readuncommitted readcommitted

repeatableread serializable

18. Whichadditionalchangeswouldmakedeadlockimpossibleinthissystem?

none(alreadyimpossible) takewritelocksbeforereadlocks

locktablesinalphabeticalorder takereadlocksbeforewritelocks

19. SupposethatSQLServerusedread/writelockswhereareadlockisalwaysgrantedprovidednowritelockhasyetbeenacquired.Inonesentence,describeaproblemwiththisapproachthatwouldbefixedbyswitchingtoSQLite’sschemewith4locktypes.

Writerscouldbewaitingforeverforatimewithnoreadlocks.

V.BigDataSystems

1. WhichfeaturesdoparallelDBsprovidethatdataflowenginesdon’t?Circleallthatapply.

declarativequeries scaletopetabytesofdata

transactions computeresultsofnon-monotonequeries

2. Thepurposeofawatermarkinastreamingsystemistocontrolwhat?

whichdatacanbeignored batchsize

howquicklyresultsmustbe numberofMapReducejobsexecuted

updatedwhennewdataarrives

3. The“shufflesort”phaseofMapReduceismostlikewhichoperationinSQL?

innerjoin groupby

outerjoin count(*)

4. ThephysicalplansinSparkSQLaddwhatelementbeyondwhatisinthelogicalplans?

wheretoacquirelocks choiceofbroadcastvsparalleljoin

wheretowritetodisk choiceofhashvsB+treeindexes

5. AdataflowpipelinewithT“reshuffle”operationscanbeexecutedwithatmosthowmanyMapReducejobs?

T-1 T

T+1 2T

Youarehiredbyawebcompanythatmakesapopularproductwithover100millionusers.

Toachievethatscale,theybuilttheirappontopofaNoSQLsystem.However,theyfound

thatitishardtoanswercomplexquestionsaboutthedatastoredinthatsystem.

TheCEOhasnowhiredyoutofixthisproblem.Shetellsyouthatyoucanhireateamof

programmerstohelpyou,butyoucannotchangeawayfromtheNoSQLsystemtheyare

alreadyusing—thatwouldbetoocostly.

Thankfully,theNoSQLsystemtheyuse,anextensiblerecordsystem(or“BigTable”),does

supportMapReduce.(Unfortunately,thesystemprovidesyouwithnostatisticsaboutthe

columnsintherecords.)Afterthinkingitover,youdecidetobuildaqueryenginethatlets

employeesexecuteSQLqueriesoverthedatabyturningSQLintoMapReducejobs.

6. Whichofthefollowingwouldyourteamnotneedtoimplementinthecodethatyou

writeforthissystem?Circleallthatapply.

parserforSQL conversionfromSQLtoRA

estimateselectivityof putRAoperatorsintogroups

conditionsinwhereclause thatrequireonlyonereshuffle

takelocksondataelements startbackupjobsforanystragglers

accordingtotwo-phaselocking

7. Whichofthefollowingwouldbethebestestimatorofthecostofaphysicalplaninthis

system?

totalCPUcycles totalCPUcyclespernode

totalnetworkI/O numberofreshuffles

8. Supposethatyouwantedtomaintainstatisticsaboutthecolumns.Yourideaistodo

thatbyaddingarecord,foreachcolumn,thatstoresitsstatistics.Whichofthefollowing

featureswouldyourequireoftheNoSQLsystem?Circleallthatapply.

row-leveltransactions generaltransactions

two-phaselocking stricttwo-phaselocking

9. DosomeNoSQLsystemsofferthatfeature?

Yes No