bad career 7

8/17/2019 Bad Career 7

1/42

How to Have a Bad CareerHow to Have a Bad Careerin Research/Academiain Research/AcademiaPre-PhD and Post-PhDPre-PhD and Post-PhD

(& How to Give a Bad Talk)(& How to Give a Bad Talk)

David Patterson

C Berkele!

"ovem#er $% '$

https://goo.gl/ChRlHT


2/42

2

Acknowled*ments & Related +ork• ,an! o these ideas came rom (ins.ired #!) Tom Anderson David C0ller Al

Davis 1en Gold#er* 2ohn Henness! 3teve 2ohnson 2ohn 40sterho0t Rand! 1at5

Bo# 3.ro0ll Carlo 3670in Bill Tet5la 8• 3t0ds Terkel Working: People talk about what they do all day and how they

feel about what they do9 ($:;


3/42

3

40tline

• Part How to Have Bad Grad 3t0dent Career and How to Avoid 4ne

• &A

• Part How to Have Bad Research Career

• Part How to Avoid a Bad Research Career

+ Richard Hamming (Turing Award forerror-detecting and error-correcting codes)video clips from “You and Your Research” (1995)

• &A

• ,! 3tor!@ Accidental Academic (? min)

• +hat +orks or ,e (? min)


4/42

4

Part @ Commandments onto Have a Bad Grad0ate Career

9 Concentrate on *ettin* *ood *rades–Postpone research involvement:

might lower GPA

–Aim for PhD class valedictorian!

Alternative@ ,aintain reasona#le *rades

–No employer cares about GPA

»Sorry, no valedictorian

–Only once I gave below B in grad course

–3 prelim courses only real grades that count

–What matters: Letters of recommendation»From 3-4 faculty & external PhDs

who have known you for 5+ years


5/42

5


9 Concentrate on *rad0atin* as ast as .ossi#le–Winner is first in class to PhD

»Only care PhD & GPA, not what you know

–Don’t spend a summer in industry: takes longer

»How could industry experience help with topic?

»Or letters of reference?

–Don’t work on large projects: takes longer

II.Have to talk to others, have to learn different areas

–Don’t do a systems PhD: takes longer

• Alternative: Your last chance to learn(mostly outside classroom)

–Considered newly “minted” when finish PhD

»No youth credit post PhD

–Judged on year of PhD vs. year of birth

–To person in 40s or 50s, 27 ≈ 29


6/42

6


9 DonIt *o to conerences–It costs money and takes time

–You’ll have plenty of time to learnthe field after graduating

• Alternative: Chance to seefirsthand what the field is like,where its going

–Talk to people in the field in thehalls as well as go to talks

–If your advisor won’t pay,then pay it yourself

»Prof. Landay paid his ownway to conferences while gradstudent

»There are student rates,can share a room


7/42

7


J9 DonIt tr0st !o0r advisor –Advisor is only interested in hisor her own career, not yours

–Advisor may try to give you work todo, which uses up your time, could⇒interfere with GPA & delay graduation

• Alternative@ Tr! tr0stin* !o0radvisor

–Primary attraction of campus vs.research lab is grad students

–Grad students reward for academic

career»Faculty career is judged by successof students

–Why not try taking advice of UCBerkeley Prof?


8/42

8

+ritin* Commandments or a Bad Career

9 Tho0 shalt not deine terms nor eK.lain an!thin*

9 Tho0 shalt re.lace = will #0ild> with=has #een #0ilt>

9 Tho0 shalt not mention draw#acks to !o0ra..roach

J9 Tho0 shalt not reerence an! .a.ers

J9 Tho0 shalt .0#lish #eore im.lementin*


9/42

9

Alternatives to Bad Pa.ers• Do o..osite o Bad Pa.er commandments

–Define terms, distinguish “will do” vs. “have done”,

–Mention drawbacks, real performance, reference other papers.

–Find related work via Google scholar…

• Lirst read 3tr0nk and +hite then ollow these ste.sM

1. 1-page paper outline, with tentative page budget/section

2. Paragraph map

»1 topic phrase/sentence per paragraph,hand drawn figures w. captions (white board & photo)

3. (Re)Write draft»Long captions/figure can contain details ~ Scientific American

»Uses Tables to contain facts that make prose dreary

4. Read aloud

5. Grammar check

»Pearson Writer ($15/year for academics) or

»MS Word - select “technical” for writing style6. Get feedback from friends and critics on draft; go to 3.

• www.cs.berkeley.edu/~pattrsn/talks/writingtips.html


10/42

10

$ Talk Commandments or a Bad Career

9 Tho0 shalt not #e neat9 Tho0 shalt not waste s.ace

9 Tho0 shalt not covet #revit!

J9 Tho0 shalt cover th! naked slides

J9 Tho0 shalt not .rint lar*e

J9 Tho0 shalt not 0se color

J9 Tho0 shalt not ill0strate

J9 Tho0 shalt not make e!e contact

N9 Tho0 shalt not ski. slides in a lon* talk

N9 Tho0 shalt not .ractice


11/42

11

Lollowin* all the commandments in Power.ointO

• We describe the philosophy and design of the control flow achine! and present the res"lts of detailed si"lations of theperforance of a single processing eleent# $ach factor is copared with the eas"red perforance of an ad%anced %on&e"ann cop"ter r"nning e'"i%alent code# (t is shown that the control flow processor copares fa%orably in the progra#

• We present a denotational seantics for a logic progra to constr"ct a control f low for the logic progra# )he control flow isdefined as an algebraic anip"lator of idepotent s"bstit"tions and it %irt"ally reflects the resol"tion ded"ctions# We alsopresent a botto*"p copilation of edi" grain cl"sters fro a fine grain control f low graph# We copare the basic bloc+and the dependence sets algoriths that partition control flow graphs into cl"sters#

• , hierarchical acro*control*flow cop"tation allows the to e-ploit the coarse grain parallelis inside a acrotas+! s"ch asa s"bro"tine or a loop! hierarchically# We "se a hierarchical definition of acrotas+s! a parallelis e-traction schee aongacrotas+s defined inside an "pper layer acrotas+! and a sched"ling schee which assigns hierarchical acrotas+s onhierarchical cl"sters#

• We apply a parallel si"lation schee to a real proble. the si"lation of a control flow architect"re! and we copare theperforance of this si"lator with that of a se'"ential one# /oreo%er! we in%estigate the effect of odeling the application onthe perforance of the si"lator# "r st"dy indicates that parallel si"lation can red"ce the e-ec"tion tie significantly ifappropriate odeling is "sed#

• We ha%e deonstrated that to achie%e the best e-ec"tion tie for a control flow progra! the n"ber of nodes within thesyste and the type of apping schee "sed are partic"larly iportant# (n addition! we obser%e that a large n"ber ofs"bsyste nodes allows ore actors to be fired conc"rrently! b"t the co"nication o%erhead in passing control to+ens totheir destination nodes ca"ses the o%erall e-ec"tion tie to increase s"bstantially#

• )he relationship between the apping schee eployed and locality effect in a progra are disc"ssed#

• /edi" grain e-ec"tion can benefit fro a higher o"tp"t bandwidth of a processor and f inally! a siple s"perscalar processorwith an iss"e rate of ten is s"fficient to e-ploit the internal parallelis of a cl"ster# ,ltho"gh the techni'"e does note-ha"sti%ely detect all possible errors! it detects nontri%ial errors with a worst*case cople-ity '"adratic to the syste sie# (tcan be a"toated and applied to systes with arbitrary loops and nondeterinis#


12/42

12

Alternatives to Bad Talks

• Do o..osite o Bad Talk commandments

• Allocate ' min0tes .er slide leave time or 70estions

• DonIt over animate

• Do dr! r0ns with riends/critics or eed#ack

–including tough audience questions

• Record a .ractice talk (video)

–

Don’t memorize speech, but have notes ready• B,@ =Givin* a irst class Qo# talkI is the sin*le most im.ortant

.art o an interview tri.9 Havin* someone know that !o0 can*ive an eKcellent talk #eore hand *reatl! increases the chanceso an invitation9 That means *ivin* *reat conerence talks9>


13/42

13

Richard Hammin* on m.ortance o Comm0nication


14/42

14

Part @ Alternatives to a Bad Grad0ate Career

• Advice rom a ver! s0ccess0l =st0dent>MRem5i Ar.aci (now +isconsin Proessor)

–Why do you think you did so well?

–Remzi: Advice you gave me first week I arrived

–What did I say?

–Remzi: 3 observations, still good advice

• =3wim or 3ink>

–“Success is determined by me (student) primarily”

–Faculty will set up opportunity, but its up to me leverage it

• =Read/learn on !o0r own>

–“Related to 1), I think you told me this as you handed me a stackof about 20 papers”

• =Teach !o0r advisor >–“I really liked this concept; go out and learn about something andthen teach the professor”

–Fast moving field, don’t expect Prof to be at forefront everywhere


15/42

15

40tline or Bad Research Career (Post-PhD)

• Part @ E Commandments or a Bad Research Career

I. Be THE leading expertII.Let Complexity Be Your Guide (Confuse Thine Enemies)

III.Never be Proven Wrong

IV.Use the Computer Scientific Method

V.Don’t be Distracted by Others (Avoid Feedback)

VI.Publishing Journal Papers IS Technology Transfer

• Part J@ Advice on Alternatives to a Bad Research Career


16/42

16

Bad Career ,ove $@ Be THS leadin* eK.ert

• nvent a new ieldO

–Make sure its slightly different

• Be the real Fone Ran*er@ DonIt work with others

–No ambiguity in credit

–Adopt the Prima Donna personality

»Prima Donna: a very temperamental person with

an inflated view of their own talent or importance• Research Hori5ons

–Never define success

–Avoid Payoffs of less than 20 years

–Stick to one topic for whole career

–Even if technology appears to leave you behind,stand by your problem


17/42

17

Bad Career ,ove '@ Fet Com.leKit! Be o0r G0ide(Con0se Thine Snemies)

• Best com.liment@=ts so com.licated canIt 0nderstand the ideas>

• Sasier to claim credit or s0#se70ent *ood ideas

–If no one understands, how can they contradict your claim?

• tIs easier to #e com.licated

–Also: to publish it must be different; N+1st incremental change• it were not 0nsim.le then how co0ld distin*0ished

collea*0es in de.artments aro0nd the world #e .ositivel!a..reciative o #oth !o0r eKtraordinar! intellect0al *ras. othe n0ances o iss0es as well as the de.th o !o0rcontri#0tion


18/42

18

Bad Career ,ove ?@ "ever #e Proven +ron*

•Avoid m.lementin*

• Avoid 0antitative SK.eriments

–If you’ve got good intuition, who needs experiments?

–Why give grist for critics’ mill?

–Plus, it takes too long to measure

• Avoid Benchmarks• ProQects whose .a!o is ' !ears *ives !o0 $: sae

!ears


19/42

19

Bad Career ,ove


20/42

20

Bad Career ,ove @DonIt #e Distracted #! 4thers (Avoid Leed#ack)

• Alwa!s dominate conversations@ 3ilence is i*norance

–Corollary: Louder is smarter

• DonIt read

• DonIt #e tainted #! interaction with 0sers ind0str!

• Reviews

–If it's simple and obvious in retrospect Reject⇒

–Quantitative results don't matter if they just show you whatyou already know Reject⇒

–Everything else Reject⇒


21/42

21

Bad Career ,ove E@P0#lishin* 2o0rnal Pa.ers 3 Technolo*! Transer

•As the leadin* scientist !o0r Qo# is to .0#lish in Qo0rnalsMits not !o0r Qo# to make !o0 the ideas .alata#le to theordinar! en*ineer

• Goin* to conerences and visitin* com.anies Q0st 0ses 0.val0a#le research time

–Travel time, having to interact with others, serve on program

committees, ...


22/42

22

Bad Career ,ove ;@+ritin* Tactics or a Bad Career

• 3t0dent .rod0ctivit! n0m#er o .a.ers

–Never ask students to implement: reduces papers

–Number of students: big is beautiful

• Fe*all! chan*e !o0r name to Aaaanderson

$

idea

<

Qo0rnal .a.ers

$E

eKtended a#stracts

E<

technical re.orts

“Publication pyramid

of

success”

• Pa.ers@ tIs 0antit! not 0alit!

–Personal Success = Length of Publication List–“The LPU (Least Publishable Unit) is Good for You”


23/42

23

40tline

•Part @ 1e! Advice or a Bad Career Pre Ph9D9

• Part @ 1e! Advice or a Bad Career Post Ph9D9

• To.ics covered in Parts Alternatives to a Bad Career

–Selecting a Problem

–Picking a Solution

–Performing the Research–Evaluating the Results

–Communicating Results

–Transferring Technology


24/42

24

4ne Alternative 3trate*! to a Bad Career • Caveats@

–From a project leader’s point of view

–Works for me; not the only way

–Primarily from researcher, computer systems perspective

• Goal is to have im.act@Change way people do Computer Science !ngineering

–Academics have bad benchmarks: number published papers

–Richard Hamming: work on important problems!

• E 3te.s

$) 3electin* a .ro#lem

') Pickin* a sol0tion

?) R0nnin* a .roQect


25/42

25

Hammin* started havin* l0nch with chemists atBell Fa#s (ater .h!sicists *ot .ri5es and let orwere .romoted)


26/42

26

$) 3electin* a Pro#lemnvent a new ield & stick to it• "oO Do =Real 3t0>@ solve .ro#lem

that others think s im.ortant–Positive Impact on CS&E

• "oO se se.arate short .roQects

–Always takes longer than expected

–Matches student lifetimes

–

Long effort in fast changing field???–Learning: Number of projects vs.calendar time

–If going to fail, better to know soon

• 3trive or m0lti-disci.linar! m0lti.leinvesti*ator .roQects

• ,atch the stren*ths and weaknesseso local environment

• ,ake s0re !o0 are eKcited eno0*h towork on it or !ears

–Prototypes help


27/42

28

') Pickin* a sol0tionFet Com.leKit! Be o0r G0ide

• "oO 1ee. thin*s sim.le 0nless a ver!

*ood reason not to –Pick innovation points carefully,and be compatible everywhere else(spend intelligence beans carefully)

–Best results are obvious in retrospect“Anyone could have thought of that”

• Com.leKit! cost is in lon*er desi*nconstr0ction test and de#0*

–Fast changing field + delaysless impressive results⇒

se the Com.0ter 3cientiic ,ethod• "oO R0n eK.eriments to discover real

.ro#lems

• se int0ition to ask 70estionsnot to answer them (40sterho0t)


28/42

29

(And Pick A Good "ameO)

Red0ced

nstr0ction

3et

Com.0tersRed0ndant

Arra! o

neK.ensive

Disks

8

"etwork

4

+orkstations


29/42

30

How do we .ick .roQect .ro#lem and sol0tion

• 3tart meetin* with ac0lt! at least $ !ear in advance todisc0ss ideas

• Track interestin* technolo*! trends over neKt -$ !earsto see i some new o..ort0nit!

–RISC: VLSI Design, Moore’s Law, 32-bit microprocessor

–RAID: 5.25” disks for PCs, low I/O performance

–NOW: Local Area Network Switches, Powerful Workstations

• Team o m0ltidisci.linar! ac0lt! to see i want tovol0nteer to take on new challen*e

• Get eed#ack on .otential .ro#lem and sol0tion romo0tsiders whose taste !o0 tr0st and iterate on vision

–Industry unlikely to compete with our project, so safer


30/42

31

Avoid Leed#ack• "oO Periodic ProQect Reviews with

40tsiders–Twice a year: 3-day retreat

–faculty, students, staff +guests

–Key piece is feedback at end

–Helps create deadlines, team spirit

–Give students chance to give

many talks / posters to interact withothers industry

• Consider mid-co0rse correction

–Fast changing field & 5 year projects⇒assumptions changed

• Pick si5e and mem#ers o team

care0ll!

–Tough personalities are hard foreveryone

–1 expert per area reduceschance of disagreement



31/42

32

DonIt #e Distracted #! 4thers• "oO 4.en Colla#orative Fa#orator!

–Avoid DSL Desert (work at home)

–Faculty, students, staff in openspace

–Aim for CommunicationandConcentration

–

Optimized meeting rooms fordiscussions and phone calls

–Kitchen, free drinks & coffee

• Accelerates researchO

–People come in more

–

Leads to spontaneous meetings–Improves 0 to 60 MPH time of newgrad students

•Hamming on importance ofOpen Space and Feedback



32/42

33

Hammin* on Doors 4.en vs9 Door Closed at Bell Fa#s


33/42

34

• People count pro"ects you finish#not the ones you start

• Successful pro"ects *o thro0*h an0n*lamoro0s hard .hase

• Desi*n is more 0n than makin* itwork

–“No winners on a losing team;

no losers on a winning team.”

–“You can quickly tell whether or notthe authors have ever builtsomething and made it work.”

• Red0ce the .roQect i its late

– “Adding people to a late projectmakes it later.”

• Linishin* a .roQect is how .eo.leac70ire taste in selectin* *ood.ro#lems indin* sim.le sol0tions


34/42

35

) Sval0atin* 0antitativel!

"ever #e Proven +ron*

• "oO !o0 canIt #e .roven wron*then !o0 canIt .rove !o0Ire ri*ht

• Re.ort in s0icient detail orothers to re.rod0ce res0lts

–can’t convince others

if they can’t get same results• Lor #etter or or worse

#enchmarks sha.e a ield

• Good ones accelerate .ro*ress

–good target for development

• Bad #enchmarks h0rt .ro*ress–help real users vs. help sales?


35/42

36

E) Transerrin* Technolo*! P0#lishin* 2o0rnal Pa.ers 3Technolo*! Transer

• "oO ,issionar! work@ =3ermons>

irst then the! read .a.ers–Selecting problem is key: “Real stuff”

»Ideally, more interest as time passes

»Change minds with believable numbers

»Prima Donnas interfere with transfer

• ,! eK.erience@ ind0str! is rel0ctantto em#race chan*e

–Howard Aiken, circa 1950:

“The problem in this business isn’t tokeep people from stealing your ideas;

itsmakingthem steal your ideas!”–Need 1 bold company (often not no. 1)to take chanceandbe successful

»RISC with Sun, RAID with (EMC, …),NOW with (Inktomi, Google…)

–Then rest of industry must follow


36/42

37

E) Transerrin* Technolo*!@ A 3tart 0.• Pros

–Everyone enjoys trying once

–Learn a lot

–Personal satisfaction: seeingyour product used by others

–Personal $$$ (potentially)

–Fame

• Cons

–Learn about business plans,sales vs. marketing, financing,personnel benefits, hiring,lawsuits …

–

Spend time doing above vs.research/development

–Only 10% of startups reallymake it


37/42

38

30mmar!@ FeaderIs Role Chan*es d0rin* ProQect


38/42

41

Concl0sion@ Alternatives to a Bad Career


–Many 5 year projects gives more chances for impact


39/42

42

,! $' Live-ear ProQects

Years Project Title (Impact) Faculty (NAE in Bold)

Students

(ACM

fellows)

1977-

1981X-Tree: A Tree-Structured Multiprocessor Despain, Patterson, Sequin 12 (2)

198-

198!"educed #nstruction Set $o%puter ("#S$-#, "#S$-##) Patterson, Ousterhout, Sequin 17 (1)

198&-

198'

SA": S%alltal n A "#S$aa *"#S$-###+

Patterson, Ousterhout 22 (1)

198-

1989

S.": S/%0olic rocessin .sin "#S$s

aa *"#S$-#+

Patterson, 3ate%an, 4il5iner,

Hodges, Katz, Ousterhout

21 (!)

1988-

1992"edundant Arra/ o5 #ne6pensie Diss ("A#D)

Katz, Ousterhout, Patterson,

Stonebraker1' (!)

199&-

1998: etor o5 orstations (#nternet $lusters) Culler, Anderson, Brewer, Patterson 2 (2)

1997-

22#"AM: #ntellient "AM

Patterson, ;u0iatoica0 (Spar, Mesos)

Patterson, 3o6, Jordan, ?osep@,

Katz, Shenker, Stoica!

27-

21&

ar >a0: arallel $o%putin >a0($o%%unication Aoidin Alorit@%s, "#S$-)

Patterson, Asanoic, Demmel, 3o6,

;euta0: Alorit@%s, Mac@ines, eople

3ranlin, Jordan, ?osep@, Katz,

Patterson, Shenker, Stoica!

212-

217

AS#"B >a0: Alorit@%s and Speciali


40/42

43

Concl0sion@ Alternatives to a Bad Career https://goo.gl/ChRlHT


–Many 5 year projects gives more chances for impact

• Do =Real 3t0 >@ make s0re !o0 are solvin* a .ro#lemthat others think is im.ortant

• 1e! is *ettin* *ood eed#ack and listenin* to it

• Taste is critical in selectin* research .ro#lems sol0tionseK.eriments and comm0nicatin* res0ltsM

–Taste acquired from feedback and completing projects

• Lac0lt! real le*ac! is .eo.le !o0 .rod0ce not .a.ers@

–Expected from reading 1974 bookWorking: People talk aboutwhat they do all day and how they feel about what they do

–Create environments that develop PhDs of whom proud

• Students are the coin o the academic realm


41/42

44

,! 3tor!@ Accidental Academic• $st colle*e *rad0ate in amil!M no C3/*rad school .lan–

Wrestler, Math major in high school and college• Accidental CFA PhD st0dent–New UCLA PhD (Jean-Loup Baer) took pity on undergrad

• +ie U ' sons in ,arried 3t0dents Ho0sin* while *radst0dent–Lost RA-ship after 4 years because grant ended–Part time at Hughes Aircraft Company 3 more years

• Accidental Berkele! Proessor –Wife forced me to call UC Berkeley CS Chair to check on application

• $

st

.roQect as Assistant Pro with an Associate Pro tooam#itio0s & no reso0rces–Took leave to DEC to rethink career in 3rd year

• Ten0re not eas! (Conerence vs9 Qo0rnal R3C too recent)

• 3till *et .a.ers reQected #! Qerks on Pro*ram Committee


42/42

+hat +orks or ,e• ,aKimi5e Personal Ha..iness vs9 Personal +ealth

• Lamil! LirstO• Passion 4.timism & Co0ra*e

– Swing for the fences vs. Bunt for singles– “Friends may come and go, but enemies accumulate”

• +innin* as Team vs9 +innin* as ndivid0al–“No losers on a winning team, no winners on a losing team”

• 3eek 40t Honest Leed#ack & Fearn Lrom t– Reliable Danger Sign: “I’m smartest person in the room”

• 4ne (Bi*) Thin* at a Time– It’s not how many projects you start;It’s how many you finish

• Have L0n@ +ork Hard Pla! Hard

bad career 7

Documents