8/27/11

1

CS61C:GreatIdeasinComputerArchitecture

(a.k.a.MachineStructures)Course Introduc-on 

Instructors:

MikeFranklin

DanGarcia

hDp://inst.eecs.berkeley.edu/~cs61c/fa11

1Fall2011‐‐Lecture#1

Agenda

•  ThinkingaboutMachineStructures

•  GreatIdeasinComputerArchitecture

•  Whatyouneedtoknowaboutthisclass

28/26/11 Fall2011‐‐Lecture#1

Agenda

•  ThinkingaboutMachineStructures

•  GreatIdeasinComputerArchitecture

•  Whatyouneedtoknowaboutthisclass

38/26/11 Fall2011‐‐Lecture#1

CS61cisNOTreallyaboutCProgramming

•  Itisaboutthehardware‐soUwareinterface– Whatdoestheprogrammerneedtoknowtoachievethehighestpossibleperformance

•  LanguageslikeCareclosertotheunderlyinghardware,unlikelanguageslikeScheme!– Allowsustotalkaboutkeyhardwarefeaturesinhigherlevelterms

– Allowsprogrammertoexplicitlyharnessunderlyinghardwareparallelismforhighperformance

8/27/11 Fall2011‐‐Lecture#1 4

OldSchoolCS61c

58/26/11 Fall2011‐‐Lecture#1

(kinda)NewSchoolCS61c(1)

6

PersonalMobileDevices

8/26/11 Fall2011‐‐Lecture#1

8/27/11

2

7

WarehouseScaleComputer

8/26/11 Fall2011‐‐Lecture#1

NewSchoolCS61c(2)Old‐SchoolMachineStructures

8

CS61c

I/OsystemProcessor

Compiler

Opera6ng

System(MacOSX)

Applica6on(ex:browser)

DigitalDesign

CircuitDesign

Instruc6onSetArchitecture

Datapath&Control

transistors

MemoryHardware

SoHware Assembler

8/26/11 Fall2011‐‐Lecture#1

New‐SchoolMachineStructures(It’sabitmorecomplicated!)

•  ParallelRequestsAssignedtocomputere.g.,Search“Katz”

•  ParallelThreadsAssignedtocoree.g.,Lookup,Ads

•  ParallelInstruccons>1instruccon@onecmee.g.,5pipelinedinstruccons

•  ParallelData>1dataitem@onecmee.g.,Addof4pairsofwords

•  HardwaredescripconsAllgatesfuncconingin

parallelatsamecme8/27/11 Fall2011‐‐Lecture#1 9

SmartPhone

WarehouseScale

Computer

So/ware        Hardware 

Harness Parallelism & Achieve High Performance 

LogicGates

Core Core…

Memory(Cache)

Input/Output

Computer

MainMemory

Core

InstrucconUnit(s) FuncconalUnit(s)

A3+B3A2+B2A1+B1A0+B0

Project1

Project2

Project3

Project4

Agenda

•  ThinkingaboutMachineStructures

•  GreatIdeasinComputerArchitecture

•  Whatyouneedtoknowaboutthisclass

108/26/11 Fall2011‐‐Lecture#1

6GreatIdeasinComputerArchitecture

1.  LayersofRepresentacon/Interpretacon2.  Moore’sLaw

3.  PrincipleofLocality/MemoryHierarchy

4.  Parallelism

5.  PerformanceMeasurement&Improvement

6.  DependabilityviaRedundancy

8/27/11 Fall2011‐‐Lecture#1 11

GreatIdea#1:LevelsofRepresentacon/Interpretacon

lw $t0,0($2)lw $t1,4($2)sw $t1,0($2)sw $t0,4($2)

HighLevelLanguageProgram(e.g.,C)

AssemblyLanguageProgram(e.g.,MIPS)

MachineLanguageProgram(MIPS)

HardwareArchitectureDescrip6on(e.g.,blockdiagrams)

Compiler 

Assembler 

Machine Interpreta4on 

temp=v[k];v[k]=v[k+1];v[k+1]=temp;

0000 1001 1100 0110 1010 1111 0101 1000 1010 1111 0101 1000 0000 1001 1100 0110 1100 0110 1010 1111 0101 1000 0000 1001 0101 1000 0000 1001 1100 0110 1010 1111

LogicCircuitDescrip6on(CircuitSchema6cDiagrams)

Architecture Implementa4on 

Anythingcanberepresentedasanumber,

i.e.,dataorinstruccons

8/27/11 12Fall2011‐‐Lecture#1

8/27/11

3

8/27/11 Fall2011‐‐Lecture#1 13

Predicts:2XTransistors/chipevery2years

GordonMooreIntelCofounderB.S.Cal1950!

#oftransistorsonan

integratedcircuit(IC)

Year

#2:Moore’sLaw JimGray’sStorageLatencyAnalogy:HowFarAwayistheData?

Registers

On Chip Cache On Board Cache

Memory

Disk

1

2 10

100

Tape /Optical Robot

10 9

10 6

Sacramento

This Campus

This Room My Head

10 min

1.5 hr

2 Years

1 min

Pluto

2,000 Years The ima

Andromeda

JimGrayTuringAwardB.S.Cal1966Ph.D.Cal1969!

GreatIdea#3:PrincipleofLocality/MemoryHierarchy

8/27/11 Spring2011‐‐Lecture#1 15

GreatIdea#4:Parallelism

8/27/11 Fall2011‐‐Lecture#1 16

8/27/11 Spring2011‐‐Lecture#1 17

Caveat:Amdahl’sLaw

GeneAmdahlComputerPioneerPh.D.Wisconsin1952!

GreatIdea#5:PerformanceMeasurementandImprovement

•  Matchingapplicacontounderlyinghardwaretoexploit:–  Locality–  Parallelism–  Specialhardwarefeatures,likespecializedinstruccons(e.g.,matrixmanipulacon)

•  Latency– Howlongtosettheproblemup– Howmuchfasterdoesitexecuteonceitgetsgoing–  Itisallabout-me to finish 

8/27/11 Fall2011‐‐Lecture#1 18

8/27/11

4

CopingwithFailures

•  4disks/server,50,000servers•  Failurerateofdisks:2%to10%/year

– Assume4%annualfailurerate

•  Onaverage,howoUendoesadiskfail?a)  1/monthb)  1/weekc)  1/dayd)  1/hour

8/27/11 Fall2011‐‐Lecture#1 19

CopingwithFailures

•  4disks/server,50,000servers•  Failurerateofdisks:2%to10%/year

– Assume4%annualfailurerate

•  Onaverage,howoUendoesadiskfail?a)  1/monthb)  1/weekc)  1/dayd)  1/hour

8/27/11 Fall2011‐‐Lecture#1 20

50,000x4=200,000disks200,000x4%=8000disksfail

365daysx24hours=8760hours

Fall2011‐‐Lecture#1

GreatIdea#6:DependabilityviaRedundancy

•  Redundancysothatafailingpiecedoesn’tmakethewholesystemfail

8/27/11 21

1+1=2 1+1=2 1+1=1

1+1=22of3agree

FAIL!

IncreasingtransistordensityreducesthecostofredundancySpring2011‐‐Lecture#1

GreatIdea#6:DependabilityviaRedundancy

•  Appliestoeverythingfromdatacenterstostoragetomemory–  Redundantdatacenterssothatcanlose1datacenterbutInternetservicestaysonline

–  Redundantdiskssothatcanlose1diskbutnotlosedata(RedundantArraysofIndependentDisks/RAID)

–  Redundantmemorybitsofsothatcanlose1bitbutnodata(ErrorCorreccngCode/ECCMemory)

8/27/11 22

Agenda

•  ThinkingaboutMachineStructures

•  GreatIdeasinComputerArchitecture

•  Whatyouneedtoknowaboutthisclass

238/26/11 Fall2011‐‐Lecture#1

Yodasays…“Alwaysinmo6onisthefuture…”

Ourschedulemaychangeslightlydependingonsomefactors.Thisincludeslectures,assignments&labs…

8/27/11

5

Hotoffthepresses

•  Duetohighstudentdemand,we’veaddedatenthseccon!!

•  It’sthesamecmeaslab105

•  Everyone(notjustthoseonthewaitlist),considermovingtothisseccon

8/27/11 Fall2011‐‐Lecture#1 25

CourseInformacon•  CourseWeb:hDp://inst.eecs.Berkeley.edu/~cs61c/•  Instructors:

–  DanGarcia,MichaelFranklin•  TeachingAssistants:

–  BrianGawalt(HeadTA),EricLiang,PaulRuan,SeanSoleyman,AnirudhTodi,andIanVonseggern

•  Textbooks:Average15pagesofreading/week(canrent!)–  PaDerson&Hennessey,Computer Organiza-on and Design,

4thEdicon(not≤3rdEdicon,notAsianversion4thedicon)–  Kernighan&Ritchie,The C Programming Language,2ndEdicon–  Barroso&Holzle,The Datacenter as a Computer, 1st Edi-on

•  Piazza:–  Everyannouncement,discussion,clarificaconhappensthere

8/27/11 Fall2011‐‐Lecture#1 27

Reminders

•  Discussionsandlabswillbeheldnextweek– SwitchingSeccons:ifyoufindanother61Cstudentwillingtoswapdiscussion(fromthePiazzathread)ANDlab,talktoyourTAs

– Partners(onlyproject2,3andperformancecompeccon)

8/27/11 Fall2011‐‐Lecture#1 28

CourseOrganizacon•  Grading

–  EPA:Effort,ParccipaconandAltruism(5%)– Homework(10%)–  Labs(5%)–  Projects(20%)

1.  ComputerInstrucconSetSimulator(C)2.  DataParallelism(Map‐ReduceonAmazonEC2)3.  PerformanceTuningofaParallelApplicacon/Matrix

Mulcplyusingcacheblocking,SIMD,MIMD(OpenMP)4.  ComputerProcessorDesign(Logisim)

– MatrixMulcplyCompecconforhonor(andEPA)– Midterm(25%):dateTBA,canbeclobbered!–  Final(35%):3‐6PMThursdayDecember15th

8/27/11 Fall2011‐‐Lecture#1 29

Tried‐and‐TrueTechnique:PeerInstruccon

•  Increasereal‐cmelearninginlecture,testunderstandingofconceptsvs.details

•  Ascompletea“segment”askmulcplechoicequescon–  1‐2minutestodecideyourself–  2minutesinpairs/triplestoreachconsensus.–  Teachothers!–  2minutediscussionofanswers,quescons,clarificacons

•  YoucangettransmiDersfromtheASUCbookstoreORyoucanuseweb>clickerappfor$10!–  We’llstartthisonMonday

8/27/11

6

EECSGradingPolicy

•  hDp://www.eecs.berkeley.edu/Policies/ugrad.grading.shtml

“AtypicalGPAforcoursesinthelowerdivisionis2.7.ThisGPAwouldresult,forexample,from17%A's,50%B's,20%C's,10%D's,and3%F's.AclasswhoseGPAfallsoutsidetherange2.5‐2.9shouldbeconsideredatypical.”

•  Fall2010:GPA2.8126%A's,47%B's,17%C's,3%D's,6%F's

•  Job/InternInterviews:Theygrillyouwithtechnicalquescons,soit’swhatyousay,notyourGPA

(New61cgivesgoodstufftosay)8/27/11 Fall2011‐‐Lecture#1 31

Fall Spring

2010 2.81 2.81

2009 2.71 2.81

2008 2.95 2.74

2007 2.67 2.76

ExtraCredit:EPA!

•  Effort– ADendingprofandTAofficehours,complecngallassignments,turninginHW0,doingreadingquizzes

•  Parccipacon– ADendinglectureandvocngusingtheclickers– Askinggreatquesconsindiscussionandlectureandmakingitmoreinteraccve

•  Altruism– HelpingothersinlaboronPiazza

•  EPA!extracreditpointshavethepotencaltobumpstudentsuptothenextgradelevel!(butactualEPA!scoresareinternal)

LatePolicy…SlipDays!•  Assignmentsdueat11:59:59PM•  Youhave3slipdaytokens(NOThourormin)

•  Everydayyourprojectorhomeworkislate(evenbyaminute)wedeductatoken

•  AUeryou’veusedupalltokens,it’s33%deductedperday.– Nocreditifmorethan3dayslate

– Saveyourtokensforprojects,worthmore!!

•  Noneedforsobstories,justuseaslipday!8/27/11 Fall2011‐‐Lecture#1 33

PolicyonAssignmentsandIndependentWork

•  Withtheexcepconoflaboratoriesandassignmentsthatexplicitlypermityoutoworkingroups,allhomeworkandprojectsaretobeYOURworkandyourworkALONE.

•  Youareencouragedtodiscussyourassignmentswithotherstudents,andextracreditwillbeassignedtostudentswhohelpothers,parccularlybyansweringquesconsonPiazza,butweexpectthatwhatyouhandinisyours.

•  ItisNOTacceptabletocopysoluconsfromotherstudents.•  ItisNOTacceptabletocopy(orstartyour)soluconsfromtheWeb.•  Wehavetoolsandmethods,developedovermanyyears,fordeteccngthis.

YouWILLbecaught,andthepenalcesWILLbesevere.•  AttheminimumNEGATIVEPOINTSfortheassignment,probablyanFinthe

course,andaleDertoyouruniversityrecorddocumencngtheincidenceofcheacng.

•  (We’vecaughtpeopleinrecentsemesters!)•  BothGiverandReceiverareequallyculpable

8/27/11 Fall2011‐‐Lecture#1 34

35Spring2011‐‐Lecture#18/27/11

ArchitectureofatypicalLecture

8/27/11 Fall2011‐‐Lecture#1 36

ADencon

Time(minutes)

10 30 35 58 60

Administrivia“Andinconclusion…”

Full

8/27/11

7

Summary•  CS61C:Learn6greatideasincomputerarchitecturetoenablehighperformanceprogrammingviaparallelism,notjustlearnC1.  LayersofRepresentacon/Interpretacon2.  Moore’sLaw3.  PrincipleofLocality/MemoryHierarchy4.  Parallelism5.  PerformanceMeasurementandImprovement

6.  DependabilityviaRedundancy

37Fall2011‐‐Lecture#18/27/11