Vldb PhD workshop 2016
-
Upload
tim-kraska -
Category
Education
-
view
190 -
download
0
Transcript of Vldb PhD workshop 2016
MethodtoMyMadnessTimKraska
TakeeverythingIsaywithagrainofsalt
Disclaimer
TopicsIworkedon
• XML• CloudComputing• Visualization• TransactionProcessing• Geo-Replication• ModernNetworks• GPUs• Probabilistic PricingModels forMarketPlaces
• DataIntegration• Crowd-Sourcing• SystemsforMachineLearning• GraphProcessing• …
TopicsIworkedon
• XML• CloudComputing• Visualization• TransactionProcessing• Geo-Replication• ModernNetworks• GPUs• Probabilistic PricingModels forMarketPlaces
• DataIntegration• Crowd-Sourcing• SystemsforMachineLearning• Graphprocessing• …
TopicsIdidNOTworkon• SkylineProcessing
TopicsIworkedon
• XML• CloudComputing• Visualization• TransactionProcessing• Geo-Replication• ModernNetworks• GPUs• Probabilistic PricingModels forMarketPlaces
• DataIntegration• Crowd-Sourcing• SystemsforMachineLearning• Graphprocessing• …
TopicsIdidNOTworkon• SkylineProcessing
Furthermore
• IwantedtodoaPhDineconomics
Furthermore
• IwantedtodoaPhDineconomics
à GladlyIdidn’t
Furthermore
• IwantedtodoaPhDineconomics
• IneverwantedtoliveintheUS
à GladlyIdidn’t
Furthermore
• IwantedtodoaPhDineconomics
• IneverwantedtoliveintheUS
à GladlyIdidn’t
à NowIamthereforalmost7years
Furthermore
• IwantedtodoaPhDineconomics
• IneverwantedtoliveintheUS
• Ineverwantedtobeaprofessor
à GladlyIdidn’t
à NowIamthereforalmost7years
Furthermore
• IwantedtodoaPhDineconomics
• IneverwantedtoliveintheUS
• Ineverwantedtobeaprofessor
à GladlyIdidn’t
à NowIamthereforalmost7years
àOK,thatoneIamstillevaluating
Furthermore
• IwantedtodoaPhDineconomics
• IneverwantedtoliveintheUS
• Ineverwantedtobeaprofessor
• Ididn’tconsiderviz-relatedworkasdeep
à GladlyIdidn’t
à NowIamthereforalmost7years
àOK,thatoneIamstillevaluating
Furthermore
• IwantedtodoaPhDineconomics
• IneverwantedtoliveintheUS
• Ineverwantedtobeaprofessor
• Ididn’tconsiderviz-relatedworkasdeep
à GladlyIdidn’t
à NowIamthereforalmost7years
àOK,thatoneIamstillevaluating
à Now,wedoalotofworkinthatarea
Furthermore
Notuncommonforanacademiccareerà Impossibletopredict
Whythoseplaces?
Outline
• ChapterI:Finding/ChangingaResearchAgenda• ChapterII:Preparinganbeingonthejobmarket• ChapterIII:Managingstudents• ChapterIV:Work/Lifebalance• ChapterV:Whybeingaprofessoris(oneof)thebestjobsintheworld
Finding(Changing)aResearchAgenda
ChapterI
• ForyourPhD• ForyourPostDoc• ForyourTenureCase• …
WhyXML
MyFirstWork
XML
Ievenjoined
Mythesiswasplannedout
Keepinmindthatwas2007
• Itwasstillnotclearifcloudisreallyathing• RememberLarryEllison:• "MaybeI'manidiot,butIhavenoideawhatanyoneistalkingabout.Whatisit?It'scompletegibberish.It'sinsane.Whenisthisidiocygoingtostop?”• “Uh,Idon'tunderstandwhatwewoulddodifferentlyinthelightofcloudcomputing,otherthanmarket...youknow,changethewordingonsomeofourads.”
….itwasthebestdecisionImade
ManmussFeste feiern wie sie fallen*(Celebratewhenthechance isthereoronemustmakehaywhenthesun shines)
*DonaldKossmann
ManmussFeste feiern wie sie fallen(Celebratewhenthechance isthereoronemustmakehaywhenthesun shines)
LessonI:Beopportunistic
TwoMainOpportunitiesItook
• Cloudcomputing
• CollaborationwithAndyvanDamatBrownà InteractiveDataExploration
ManyothersIrejected(sometimesforgood,sometimesstupidly,e.g.,Spark)
Currentopportunisticopportunities
• Modernhardware(GPUs,fastnetworks,newtypesofmemory,…)• Cyborgidea(seeAnand Rajaraman’s keynote)• Deeplearning???• Betterintegrationofhumans• …
Thatdoesnotmeanyoushouldrunaftereverynewcarrot
LessonII:Focusontheffuunnddaammeennttaall iissssuueess
Example:UnicornHardware
NVM
FFiirrssttRRoouunnddooffRReesseeaarrcchh
Fundamentalresearchonhowthehardwarewouldchangetheworld
SSeeccoonnddRRoouunnddooffRReesseeaarrcchh
DB+NVM+CrowdDB+NVM+SkylineDB+NVM+Cloud+Crowd
FFiirrssttRRoouunndd::RRaaddiiooaaccttiivviittyy
SSeeccoonnddRRoouunndd::RRaaddiiooaaccttiivviittyy
BDS3Howtousecloud-basedstorageasadisk
SIGMOD2008
MMyyoowwnneexxaammpplleess
Page
1
Page
2
Page
3
Page
4
Page
5
Page
N
BuildingaDatabaseonS3
RecordManager
PageManager
Application
Client 1/EC2
RecordManager
PageManager
Application
ClientM/EC2
........
........
........
CouldbeexecutedonEC2orcompletelyonthe
client
Page
6
BDS3Howtousecloud-basedstorageasadisk
Fundamentalquestion:Whatdoesconsistencymeanforcostandscalability
SIGMOD2008
ConsistencyRationingVLDB2009
BasicIdeausedbyseveralstartups(sonotthatstupidafterall)
MMyyoowwnneexxaammpplleess
LessonIII:Donotoverthinkandstartquicklyandwithwhatyouknow
SSttaarrttWWiitthhWWhhaattYYoouuKKnnooww
Examples•SQLforCrowdsourcing•UseS3asadiskfordatabases
Doitfast
Asaruleformystudents:foranewtopicourgoalisoftentowritethefirstinitialpaperin3-9month• Thiscanbeavisionpaperforaworkshop,CIDR,etc.orinrarecasesafullpaper
àHelpstoarticulatethegoalsàHelpstogetearlyfeedback
Whatreactiondoyouwant?
1) Yeah,thatmakessense2) Peoplearguing,fightingagainstyou,….
ReactionsonCrowdDB:
Best IdeaEver Youdidn’tciteme Whataretheysmoking inBerkeley
Whatreactiondoyouwant?
1) Yeah,thatmakessense2) Peoplearguing,fightingagainstyou,….
ReactionsonCrowdDB:
Best IdeaEver Youdidn’tciteme Whataretheysmoking inBerkeley
• Ifyoujumpstartanewproject,itwillnotbeperfectanditwilltaketimetorefineit.• Example:
RecalllessonII:Focusontheffuunnddaammeennttaall iissssuueess
CrowdDb
MetaD
ata
Statistic
sParser
Optimizer
Executor
FileAccessMethods
WorkerRelationshipManager
UICreation
UITemplateManager
FormEditor
HITManager
Disk2
Disk1
ResultCrowdSQLSQL
44
Howtointegratecrowdworkers inqueryprocessing
Fundamentalquestion:Whatisthequerysemanticsintheopenworld?
BestpaperawardatICDE2013
VLDB2011
Howtousecrowdsourcingfordatacleaning
SeveralpapersatVLDBandSIGMOD
LLeessssoonnIIVV::SSiimmpplliicciittyyiissccoorreeffoorraarreesseeaarrcchhaaggeennddaa• Youshouldbeabletodescribetheprobleminlessthan30seconds.• Ideally,inasinglesentence.
LLeessssoonnVV::TThheerreeiissaattiimmeettoorreeaaddppaappeerrssaannddrreellaatteeddwwoorrkkbbuuttnnoottiinntthheebbeeggiinnnniinngg
LLeessssoonnVVII::DDoonnoottuunnddeerr--eessttiimmaatteeooppeenn--ssoouurrcceeaanndd““mmaarrkkeettiinngg””
LLeessssoonnVVIIII::PPaappeerrssaarreennootteevveerryytthhiinngg,,bbuuttaanneecceessssiittyyDisclaimer:Verycontroversial
TheadviseIgotfrommyadvisor,andhegotfromhisadvisor,is:• TheonlythingthatreallycountsareSIGMODand
VLDBpaper• Havingacertainamountisanecessitybutnot
sufficienttolandagoodacademicjob
Goalwithmystudents:1top-tierpaperperyearstartingfromyear2.
LLeessssoonnVVIIIIII::DDoowwhhaattiissffuunn
Preparingandbeingonthejobmarket
ChapterII
• ForyourPhD• ForyourPostDoc• ForyourProfessorship• Partiallyalsoforlandingyourdreamindustryjob
UnderstandtheAcademicMarket
•USA• FixcyclestartinginDecember• Tenure-trackmodel• Differentlong-termresources• Moreflexebility
• Europe(Germany,Switzerland,UK)• Ad-hod• Veryslowprocess• Extremelyhard• Faketenure-trackmodel
Icanonlycommentonthosetwomarkets
Decidingwheretoapply?
Hardconstraints
LessonIX:Donotunder-estimatetheimportanceofnetworking
• Itisimportantthatyouknowpeopleinthecommunity• Evenmoreimportantisit,thattherightpeopleknowyou
NetworkingTips• 30s/2min/5minrule• Donotspendthetimewithyourcolleaguesataconference• Visitcompanies/universitiesoutsideoftheconferencesandasktogiveatalk• Activelyintroduceyourself(theeasiestistoaskaquestionafteratalk)• Identifythepeople,whoshouldknowyou• Professorsatcertainplaces,companies,etc.• PhDstudentsatcertainuniversities…(yourPhDfriendsmightbecomeyourcollegues)
Don’ttrytohardà itshouldcomenaturalandwithrealinterestItalsohelpstobefriendswithveryoutgoingpeople.
MostConnectionsAreMadeOutsidetheVenue
LessonX:Rankingsmatter,butonlytoacertaindegreeIntheUS,everythingiscenteredaroundtheUSNewsRankingWhydoesitmatter:students!!!!But,imho:
LessonX:Rankingsmatter,butonlytoacertaindegreeIntheUS,everythingiscenteredaroundtheUSNewsRankingWhydoesitmatter:students!!!!But,imho:1. Yourresearchgroupcountsmorethanthe
institution
LessonX:Rankingsmatter,butonlytoacertaindegreeIntheUS,everythingiscenteredaroundtheUSNewsRankingWhydoesitmatter:students!!!!But,imho:1. Yourresearchgroupcountsmorethanthe
institution2. Youradvisor countsmorethantheresearch
group
LessonX:Rankingsmatter,butonlytoacertaindegreeIntheUS,everythingiscenteredaroundtheUSNewsRankingWhydoesitmatter:students!!!!But,imho:1. Yourresearchgroupcountsmorethanyour
institution2. Youradvisor countsmorethanyourresearch
group3. Yourhappiness countsmorethanyouradvisor
LessonX:Rankingsmatter,butonlytoacertaindegreeIntheUS,everythingiscenteredaroundtheUSNewsRankingWhydoesitmatter:students!!!!But,imho:1. Yourresearchgroupcountsmorethanyour
institution2. Youradvisor countsmorethanyourresearch
group3. Yourhappiness countsmorethanyouradvisorMyexample:
vs.
LessonXI:LookfortherightCollaborators• You,asasingleperson,donotscale• Onewaytoscaleis,havealotofcollaboratorsyoucantrust• Important:Ithastobeamutualcollaboration• Ifyouevaluateaplace,considerpeoplewhoyoucouldcollaboratewith.Especially,lookfor• Theycollaborations inthepast• Trytotalktobothparties• Doyougetalong?Doesitclick?
• Butkeepinmind:intheend successfulcollaborationshappenbecauseyoulikeeachotherandtheyarenormallynevertheonesyouplannedtohave.
Students,students,students,…
ChapterIII
• Managingbachelor students• Managingmaster students• ManagingPhD students• Managinginterns
LessonXII:Studentsareyourmostvaluableasset• PhDstudents,Masterstudents,bachelorstudents,…• Carefullyselectthem• Spendtimeonthepromisingonesandlessontheothers (note,thatthisisverycontroversial)• Trytofindyourbalancebetweenhands-onandhands-off
Dangeroftoohands-on
• Theymightgetfrustrated• Ittakesupalotofyourtime
Dangeroftoohands-off• Ittakesmuchlonger• itendsupinengineeringratherthanresearch• ittakesmoreiterations(e.g.,re-submissions)• Thetopicderails• Ifthetopicderailstoomuch,itmightnotfitwithyourfundinganymore
Example:Letsassumeyougiveastudentatasktodesignanactionfigurewaterpistol
AcademicJobRamp-Up
Mysuggestion:alwayspostponetostartyouracademicjobinspring notfall.• YoucanbeinvolvedinthePhDhiring(happeninginspring).• Youminimizethetimewithoutstudents(fallsemester)• Youcanfinishsomeofyourpreviousworkwithoutstartingthetenureclock orsimplytaketimeofftogetreadyforbeingaprofessor
Work/LifeBalance
ChapterIV
NNoonn--LLeessssoonn::IIssttiillllddoonnoottkknnoowwhhoowwttooddooiitt
Whybeingaprofessoris(oneof)thebestjobs
ChapterIV
Flexibility
Freedom
Travel
http://kingofwallpapers.com/travel/travel-022.jpg
Community
Startups
ThankYouTimKraska<[email protected]>
Wedohaveseveralintern(paid6-12month),PhD,andpostdocpositionsavailable