Computer Architecture

1
Computer Architecture Dr. W.
Intro
2
Overview

• Goal of this course
• break open the box
• examine the guts of computers and system software
• reveal the secrets of computers

3
Advances in Technology

• Processor
• logic capacity about 30 per year
• clock rate about 20 per year
• Memory
• DRAM capacity about 60 per year (4x every 3
  years)
• Memory speed about 10 per year
• Cost per bit improves about 25 per year
• Disk
• capacity about 60 per year

4
Motivation

• Embedded Systems
• automotive electronics
• high-end BMW has 88 microprocessors on 3
  different busses
• drive-by-wire, anti-lock brakes, active
  suspension
• avionics
• fly-by-wire
• consumer electronics
• PDAs, cell phones, set-top boxes, stereo
  equipment
• scientific equipment
• Human Genome Project, measurement devices

5
Moore's Law

• Moores Law the complexity of integrated
  circuits will double every 12, 18, 24 months
• (Intel co-founder and chairman emeritus Gordon
  Moore.
• Excerpts below from a July 10, 2002 interview )
• Process technology able to put more and more
  functionality on the same chip as the cpu

6
History

• J. Presper Eckert and John Mauchly build first
  electronic general-purpose computer at Moore
  School of UPenn during WWII
• ENIAC - Electronic Numerical Integrator and
  Calculator
• funded by US Army to calculate artillery firing
  tables
• 80 feet long by 8.5 feet high
• 18,000 vacuum tubes
• 20 10-digit registers (each 2 feet long)
• 1900 additions per second (0.19 MIPS machine)
• programmed by plugging cables and setting switches

7
Generations

• First generation
• abacus, punch cards for textile machines
• Second generation (1940 - 1960)
• ENIAC
• Whirlwind Project at MIT
• 2K magnetic core memory (popular for 30 years,
  NASA)
• 16 bit words
• vacuum tubes
• 100 times larger than todays machines
• 1/10,000 speed of todays machines
• programmed in machine code (1s and 0s)

8
Generations

• Third generation (1960 -1968)
• Microprogramming introduced
• IBM 360 family
• Early work with windows, pointing devices,
  networking (XEROX PARC)
• Programming in FORTRAN, COBOL, Basic
• Fourth generation (1969 -1977)
• Minicomputers
• First microprocessors
• UNIX
• Assembler, C, Pascal, Modula, Smalltalk

9
Generations

• Fifth generation (1978 - ?)
• VLSI (Very Large Scale Integration)
• DOS, CPM, MacOS, Windows NT, OS/2, Linux
• ADA, C, Java, HTML

10
Big Picture
Microsoft Office
Operating Systems
Netscape
Compilers/ Assemblers
Processor
Caches
System bus
11
Performance

• Memory model has impact on performance
• In 1960s 1970s memory was very expensive
• fast programs were small programs
• Today caches are relatively small (64-256Kbytes)
• fast programs are small programs
• To make a program faster we need to find the
  bottleneck
• Amdahls Law optimize the common case (well
  revisit later)
• How fast should a desktop computer be to make web
  surfing faster?

12
Classic Computer Organization
Computer
13
Computer Organization

• Comprised of 4 parts
• Input - mouse, keyboard, microphone, disk
• Output - display, disk, sound
• Memory - main memory, cache, video ram
• CPU
• ALU
• control logic
• Parts are connected by busses
• Bus
• address
• data
• control
• common busses PCI, SCSI, USB, serial, ethernet,
  Firewire

14
Memory

• Storage for
• instructions
• data
• Recipe box analogy
• ingredients (data)
• recipe (instructions)

15
Memory

• DRAM - dynamic random access memory
• very dense (cheap) because memory cell is simple
• need to periodically refresh or contents lost
• commonly used as main memory
• SRAM - static random access memory
• cell is more complicated (expensive)
• refresh not required
• cache memory
• Hard disks
• cheap for the amount of storage
• 10 gigabyte of RAM vs. 10 gigabyte disk
• ram access 50ns disk access 5ms (5000ns)

16
DRAM capacity
17
Where are we now?

• "... global DRAM demand is gathering steam
  following the semiconductor industry's worst-ever
  slowdown. Prices of standard 128Mbit DRAM have
  recovered significantly from about 1 last year,
  analysts said Infineon expects that production
  at Winbond will begin with 256Mbit DDR memory
  chips, with a potential future migration to
  512Mbit components. " 03/11/02
• http//www.ebnonline.com/story/OEG20020311S0016

18
Architecture

• Two main architectures Von Neumann and Harvard
• John Von Neumann from Princeton
• Howard Aiken from Harvard
• Von Neumann architecture
• combines data and instructions in same physical
  memory
• Harvard architecture
• data and instructions each have their own
  physical memory

19
Architecture

• Most contemporary machines based on Von Neumann
• Von Neumann bottleneck
• memory bandwidth
• data and address signals use same bus

20
Instruction Set Architecture (ISA)

• A very important abstraction
• interface between hardware and low-level software
• standardizes instructions, machine language bit
  patterns, ...
• advantage different implementations of the same
  architecture
• disadvantage sometimes prevents using new
  innovations
• consider Intels 80x86 ISA
• same software runs on Pentium, 80486, 80386,
  80286, 8086
• Modern instruction set architectures
• 80x86/Pentium/P6, PowerPC, DEC Alpha, MIPS,
  SPARC, HP

21
Workstation performance
22

• http//www.anandtech.com/

23
Software

• Computer is a programmable machine
• changing software changes its functionality
• able to act like other machines (JavaVM, Virtual
  PC)
• Early programming in machine code
• Wrote a program to output machine code from
  assembly
• Wrote a program to output assembly from a higher
  level language (e.g. C)
• Wrote a program to output C from an even higher
  level language
• Ex Janus is a language designed for distributed
  constraint programming. The tool jc compiles
  Janus programs down to C code.

24
Combining hardware and software

• Functional specification find sum of 2 and 3
• Need to translate above into a representation
  that computer can handle
• can have many different translations (not
  unique)
• translation is called software design
• result is a computer program
• Still need to translate to get to machine language

25
Application Software (User Program)
26
Compilers/Assemblers

• A compiler is a tool for translating programs
• One target is assembly language
• At this point, program is machine-dependent
• Assembler takes assembly language and outputs
  machine language
• binary code (1s and 0s)

27
Linker/Loader

• All variables and data structures
• must reside in memory
• each needs an address
• Addresses
• which address should we use
• absolute vs relocatable
• assembler generates relocatable code
• How do we use existing code without typing it in
  every time?
• Linker/loader
• links program modules together
• resolves addresses
• At this point we have a computer program on the
  hard drive

28
Running the program

• Program copied from hard drive to main memory
• Next to cache
• Finally to instruction register