Chapter 1

Computer Abstractions and Technology

#$%$ &ntroduction

The Computer Revolution

Progress in computer technology

Underpinned by Moores Law Computers in automobiles Cell phones Human genome project orld ide eb !earch "ngines
Chapter 1 Computer Abstractions and Technology 2

Makes novel applications feasible

Computers are pervasive

Classes of Computers

'esktop computers

(eneral purpose) variety of software !ubject to cost*performance tradeoff +etwork based High capacity) performance) reliability ,ange from small servers to building si-ed Hidden as components of systems !tringent power*performance*cost constraints

!erver computers

"mbedded computers

Chapter 1 Computer Abstractions and Technology 3

The Processor Market

Chapter 1 Computer Abstractions and Technology 4

What You Will Learn

How programs are translated into the machine language

.nd how the hardware e/ecutes them

0he hardware*software interface hat determines program performance

.nd how it can be improved

How hardware designers improve performance hat is parallel processing

Chapter 1 Computer Abstractions and Technology 5

Understanding Performance

.lgorithm

'etermines number of operations e/ecuted 'etermine number of machine instructions e/ecuted per operation 'etermine how fast instructions are e/ecuted 'etermines how fast &*1 operations are e/ecuted Chapter 1 Computer Abstractions and Technology 6

Programming language) compiler) architecture

Processor and memory system

&*1 system 2including 1!3

#$%7 8elow 9our Program

Below Your Program

.pplication software

ritten in high4level language Compiler5 translates HLL code to machine code 1perating !ystem5 service code

!ystem software

Handling input*output Managing memory and storage !cheduling tasks 6 sharing resources

Hardware

Processor) memory) &*1 controllers

Chapter 1 Computer Abstractions and Technology 7

Levels of Program Code

High4level language

Level of abstraction closer to problem domain Provides for productivity and portability 0e/tual representation of instructions 8inary digits 2bits3 "ncoded instructions and data Chapter 1 Computer Abstractions and Technology 8

.ssembly language

Hardware representation

#$%< Under the Covers

Components of a Computer
The BIG Picture

!ame components for all kinds of computer

'esktop) server) embedded User4interface devices

&nput*output includes

'isplay) keyboard) mouse Hard disk) C'*':') flash ;or communicating with other computers

!torage devices

+etwork adapters

Chapter 1 Computer Abstractions and Technology

natom! of a Computer
1utput device

+etwork cable

&nput device

&nput device

Chapter 1 Computer Abstractions and Technology 1!

natom! of a Mouse

1ptical mouse

L"' illuminates desktop !mall low4res camera 8asic image processor

Looks for /) y movement

8uttons 6 wheel

!upersedes roller4ball mechanical mouse

Chapter 1 Computer Abstractions and Technology 11

Through the Looking "lass

LC' screen5 picture elements 2pi/els3

Mirrors content of frame buffer memory

Chapter 1 Computer Abstractions and Technology 12

#pening the Bo$

Chapter 1 Computer Abstractions and Technology 13

%nside the Processor &CPU'

'atapath5 performs operations on data Control5 se=uences datapath) memory) %%% Cache memory

!mall fast !,.M memory for immediate access to data

Chapter 1 Computer Abstractions and Technology 14

%nside the Processor

.M' 8arcelona5 > processor cores

Chapter 1 Computer Abstractions and Technology 15

(stractions
The BIG Picture

.bstraction helps us deal with comple/ity

Hide lower4level detail 0he hardware*software interface 0he &!. plus system software interface 0he details underlying and interface

&nstruction set architecture 2&!.3

.pplication binary interface

&mplementation

Chapter 1 Computer Abstractions and Technology 16

)afe Place for *ata

:olatile main memory

Loses instructions and data when power off Magnetic disk ;lash memory 1ptical disk 2C',1M) ':'3

+on4volatile secondary memory

Chapter 1 Computer Abstractions and Technology 17

+etworks

Communication and resource sharing Local area network 2L.+35 "thernet

ithin a building

ide area network 2 .+5 the &nternet ireless network5 i;i) 8luetooth

Chapter 1 Computer Abstractions and Technology 18

Technolog! Trends

"lectronics technology continues to evolve

&ncreased capacity and performance ,educed cost

0echnology :acuum tube 0ransistor &ntegrated circuit 2&C3 :ery large scale &C 2:L!&3 Ultra large scale &C

',.M capacity

9ear $?@$ $?A@ $?B@ $??@ 7CC@

,elative performance*cost $ <@ ?CC 7)>CC)CCC A)7CC)CCC)CCC

Chapter 1 Computer Abstractions and Technology 1

#$%> Performance

*efining Performance

hich airplane has the best performanceD

Chapter 1 Computer Abstractions and Technology 2!

Response Time and Throughput

,esponse time

How long it takes to do a task 0otal work done per unit time

0hroughput

e%g%) tasks*transactions*E per hour

How are response time and throughput affected by

,eplacing the processor with a faster versionD .dding more processorsD

ell focus on response time for nowE

Chapter 1 Computer Abstractions and Technology 21

Relative Performance

'efine Performance F $*"/ecution 0ime GH is n time faster than 9I

Performanc e H Performanc e 9 = "/ecution time 9 "/ecution time H = n

"/ample5 time taken to run a program

$Cs on .) $@s on 8 "/ecution 0ime8 * "/ecution 0ime. F $@s * $Cs F $%@ !o . is $%@ times faster than 8
Chapter 1 Computer Abstractions and Technology 22

Measuring ,$ecution Time

"lapsed time

0otal response time) including all aspects

Processing) &*1) 1! overhead) idle time

'etermines system performance 0ime spent processing a given job

CPU time

'iscounts &*1 time) other jobs shares

Comprises user CPU time and system CPU time 'ifferent programs are affected differently by CPU and system performance

Chapter 1 Computer Abstractions and Technology 23

CPU Clocking

1peration of digital hardware governed by a constant4rate clock

Clock period

Clock 2cycles3 'ata transfer and computation Update state

Clock period5 duration of a clock cycle

e%g%) 7@Cps F C%7@ns F 7@CJ$CK$7s

Clock fre=uency 2rate35 cycles per second

?He%g%) >%C(HF >CCCMHF >%CJ$C Chapter 1 Computer Abstractions and Technology 24

CPU Time
CPU 0ime = CPU Clock Cycles Clock Cycle 0ime CPU Clock Cycles = Clock ,ate

Performance improved by

,educing number of clock cycles &ncreasing clock rate Hardware designer must often trade off clock rate against cycle count
Chapter 1 Computer Abstractions and Technology 25

CPU Time ,$ample

Computer .5 7(H- clock) $Cs CPU time 'esigning Computer 8

.im for As CPU time Can do faster clock) but causes $%7 J clock cycles

How fast must Computer 8 clock beD Clock Cycles8 $%7 Clock Cycles . Clock ,ate8 = = CPU 0ime 8 As Clock Cycles . = CPU 0ime . Clock ,ate . = $Cs 7(H- = 7C $C ? $%7 7C $C ? 7> $C ? Clock ,ate8 = = = >(HAs As Chapter 1 Computer Abstractions and Technology 26

%nstruction Count and CP%

Clock Cycles = &nstruction Count Cycles per &nstruction CPU 0ime = &nstruction Count CP& Clock Cycle 0ime &nstruction Count CP& = Clock ,ate

&nstruction Count for a program

'etermined by program) &!. and compiler 'etermined by CPU hardware &f different instructions have different CP&

.verage cycles per instruction

Chapter 1 Computer Abstractions and Technology 27

.verage CP& affected by instruction mi/

CP% ,$ample

Computer .5 Cycle 0ime F 7@Cps) CP& F 7%C Computer 85 Cycle 0ime F @CCps) CP& F $%7 !ame &!. hich is faster) and by how muchD
CPU 0ime . 8 = &nstruction Count CP& Cycle 0ime . . = & 7%C 7@Cps = & @CCps . is fasterE = &nstruction Count CP& Cycle 0ime 8 8 = & $%7 @CCps = & ACCps

CPU 0ime CPU 0ime

8 = & ACCps = $%7 Eby this much CPU 0ime & @CCps . Chapter 1 Computer Abstractions and Technology 28

CP% in More *etail

&f different instruction classes take different numbers of cycles

Clock Cycles = 2CP&i &nstruction Count i 3
i=$ n

eighted average CP&

n Clock Cycles &nstruction Count i CP& = = CP&i &nstruction Count i=$ &nstruction Count

,elative fre=uency

Chapter 1 Computer Abstractions and Technology 2

CP% ,$ample

.lternative compiled code se=uences using instructions in classes .) 8) C

Class CP& for class &C in se=uence $ &C in se=uence 7 . $ 7 >

8 7 $ $

C < 7 $

!e=uence $5 &C F @

!e=uence 75 &C F A

Clock Cycles Clock Cycles F 7J$ L $J7 L 7J< F >J$ L $J7 L $J< F $C F? .vg% CP& F ?*A F $%@ .vg% CP& F $C*@ F 7%C Chapter 1 Computer Abstractions and Technology 3!

Performance )ummar!
The BIG Picture

&nstructions Clock cycles !econds CPU 0ime = Program &nstruction Clock cycle

Performance depends on

.lgorithm5 affects &C) possibly CP& Programming language5 affects &C) CP& Compiler5 affects &C) CP& &nstruction set architecture5 affects &C) CP&) 0c
Chapter 1 Computer Abstractions and Technology 31

#$%@ 0he Power

Power Trends

all

&n CM1! &C technology

Power = Capacitive load :oltage 7 ;re=uency
Chapter 1 Computer Abstractions and Technology 32
J<C @: M $: J$CCC

Reducing Power

!uppose a new CPU has

N@O of capacitive load of old CPU $@O voltage and $@O fre=uency reduction

Pnew Cold C%N@ 2:old C%N@37 ;old C%N@ > = = C%N@ = C%@7 7 Pold Cold :old ;old

0he power wall

e cant reduce voltage further e cant remove more heat

Chapter 1 Computer Abstractions and Technology 33

How else can we improve performanceD

#$%A 0he !ea Change5 0he !witch to Multiprocessors

Uniprocessor Performance

Constrained by power) instruction4level parallelism) memory latency

Chapter 1 Computer Abstractions and Technology 34

Multiprocessors

Multicore microprocessors

More than one processor per chip Compare with instruction level parallelism

,e=uires e/plicitly parallel programming

Hardware e/ecutes multiple instructions at once Hidden from the programmer

Hard to do
Programming for performance Load balancing 1ptimi-ing communication and synchroni-ation Chapter 1 Computer Abstractions and Technology 35

#$%B ,eal !tuff5 0he .M' 1pteron H>

Manufacturing %Cs

9ield5 proportion of working dies per wafer

Chapter 1 Computer Abstractions and Technology 36

M* #pteron -. Wafer

H75 <CCmm wafer) $$B chips) ?Cnm technology H>5 >@nm technology

Chapter 1 Computer Abstractions and Technology 37

%ntegrated Circuit Cost

Cost per wafer Cost per die = 'ies per wafer 9ield 'ies per wafer afer area 'ie area $ 9ield = 2$+ 2'efects per area 'ie area*7337

+onlinear relation to area and defect rate

afer cost and area are fi/ed 'efect rate determined by manufacturing process 'ie area determined by architecture and circuit design Chapter 1 Computer Abstractions and Technology 38

)P,C CPU Benchmark

Programs used to measure performance

!upposedly typical of actual workload 'evelops benchmarks for CPU) &*1) eb) E

!tandard Performance "valuation Corp 2!P"C3

!P"C CPU7CCA

"lapsed time to e/ecute a selection of programs

+egligible &*1) so focuses on CPU performance

+ormali-e relative to reference machine !ummari-e as geometric mean of performance ratios

C&+07CCA 2integer3 and C;P7CCA 2floating4point3

n

i =$ Chapter 1 Computer Abstractions and Technology 3

"/ecution time ratio

C%+T.//0 for #pteron -1 .230

+ame perl b-ip7 gcc mcf go hmmer sjeng lib=uantum h7A>avc omnetpp astar /alancbmk 'escription &nterpreted string processing 8lock4sorting compression (+U C Compiler Combinatorial optimi-ation (o game 2.&3 !earch gene se=uence Chess game 2.&3 Puantum computer simulation :ideo compression 'iscrete event simulation (ames*path finding HML parsing &CJ$C? 7)$$N 7)<N? $)C@C <<A $)A@N 7)BN< 7)$BA $)A7< <)$C7 @NB $)CN7 $)C@N CP& C%B@ C%N@ $%B7 $C%CC $%C? C%NC C%?A $%A$ C%NC 7%?> $%B? 7%BC 0c 2ns3 C%>C C%>C C%>B C%>C C%>C C%>C C%>N C%>C C%>C C%>C C%>C C%>C "/ec time A<B N$B 7> $)<>@ B7$ N?C <B $)C>B ??< A?C BB< $)$>< ,ef time ?)BBB ?)A@C N)C@C ?)$7C $C)>?C ?)<<C $7)$CC 7C)B7C 77)$<C A)7@C B)C7C A)?CC $$%B !P"Cratio $@%< $$%N $$%$ A%N $>%A $C%@ $>%@ $?%N 77%< ?%$ ?%$ A%C

(eometric mean

High cache miss rates

Chapter 1 Computer Abstractions and Technology 4!

)P,C Power Benchmark

Power consumption of server at different workload levels

Performance5 ssjQops*sec Power5 atts 2Roules*sec3

$C $C att = ssjQops i poweri i=C i=C

1verall ssjQops per

Chapter 1 Computer Abstractions and Technology 41

)P,Cpower4ss5.//6 for -1
0arget Load O $CCO ?CO NCO BCO ACO @CO >CO <CO 7CO $CO CO 1verall sum SssjQops* Spower Performance 2ssjQops*sec3 7<$)NAB 7$$)7N7 $N@)NC< $A<)>7B $>C)$AC $$N)<7> ?7C)<@ BC)@CC >B)$7A 7<)CAA C $)7N<)@?C .verage Power 2 atts3 7?@ 7NA 7B@ 7A@ 7@A 7>A 7<< 777 7CA $NC $>$ 7)AC@ >?<

Chapter 1 Computer Abstractions and Technology 42

#$%N ;allacies and Pitfalls

Pitfall7 mdahl8s Law

&mproving an aspect of a computer and e/pecting a proportional improvement in overall performance

0improved 0affected = + 0unaffected improvemen t factor

"/ample5 multiply accounts for NCs*$CCs

How much improvement in multiply performance to get @J overallD NC Cant be doneT 7C = + 7C n Chapter 1 Computer Abstractions and Technology 43

Corollary5 make the common case fast

9allac!7 Low Power at %dle

Look back at H> power benchmark

.t $CCO load5 7?@ .t @CO load5 7>A 2N<O3 .t $CO load5 $NC 2A$O3 Mostly operates at $CO K @CO load .t $CCO load less than $O of the time

(oogle data center

Consider designing processors to make power proportional to load

Chapter 1 Computer Abstractions and Technology 44

Pitfall7 M%P) as a Performance Metric

M&P!5 Millions of &nstructions Per !econd

'oesnt account for

'ifferences in &!.s between computers 'ifferences in comple/ity between instructions

&nstruction count M&P! = "/ecution time $C A &nstruction count Clock rate = = A &nstruction count CP& CP& $C A $C Clock rate

CP& varies between programs on a given CPU

Chapter 1 Computer Abstractions and Technology 45

#$%? Concluding ,emarks

Concluding Remarks

Cost*performance is improving

'ue to underlying technology development &n both hardware and software 0he hardware*software interface

Hierarchical layers of abstraction

&nstruction set architecture

"/ecution time5 the best performance measure Power is a limiting factor

Use parallelism to improve performance

Chapter 1 Computer Abstractions and Technology 46