The First 31 Years of the Internet -- An Insider

1
The First 31 Years of the Internet -- An
Insiders View.Bob BradenUSC Information
Sciences Institute Oct 4, 2002
2
2

• Outline
• A. Historical Overview
• 1961 - 1968 Pre-history
• 1969 - 1973 ARPAnet research period
• 1974 - 1983 Internet Research Period
• 1984 - 1990 Academic Internet Period
• 1990 - ?? Commercial (and Popular) Internet
• B. The Internet Architecture
• Protocols
• Principles
• C. Conclusions and Challenges

3

• 1961 - 1968 Pre-History
• Computers
• Mainframes (mini-computers beginning).
• (Discrete) transistors had just replaced tubes.
• Data unit was a word of 8, 9, 12, 15, 18, 24, 32,
  36, 48, 60 ... bits. Uniform 8-bit bytes were
  introduced in 1965.
• Computer communication?
• Why? What would my IBM mainframe SAY to your
  Univac mainframe???
• Too expensive to lease telephone lines.
• The context -- Flower children, JFK LBJ, the
  Vietnam War ...

4

• 1961 - 1968 Pre-History of Computer
  Communication
• MIT
• J. C. R. Licklider Memos on "Galactic Network",
  1962
• L. Roberts Plan to build a computer network,
  1967
• Rand Corp.
• Paul Baran invented packet switching as a concept
• NPL (UK)
• o Donald Davies Roger Scantlebury wrote paper
  on a packet-switching network, 1967
• IRIA (France)o Louis Pouzin Cyclades network

5

• 1969 - 1973 ARPAnet Research Period
• ARPA Advanced Research Projects Agency
• A DoD agency formed post-Sputnik to fund
  long-range research of ultimate importance to the
  military.
• The ARPAnet
• ARPA selected Bolt, Beranek, and Newman (BBN) to
  design, build, and operate a prototype
  packet-switching network, the ARPAnet.
• Host Computers
• ARPAnet software was developed by university
  researchers, funded by ARPA.

6

• Packet Switching
• Break data into small packets, each of which
  carries its own address. Packets of different
  users are interlaced (multiplexed) on e.g. a
  leased telephone line.

• Build a network of "packet switches"
• Each switch receives, stores, and forwards
  packets.
• Packets are forwarded hop-by-hop, from any
  source host computer to any destination host
  computer.

7

• Packet-Switching (cont)
• ARPANET packet switches ("IMPs")
• Minicomputers
• High-speed leased lines (56Kbps)
• Distributed adaptive routing algorithm
• The ARPAnet grew from 10 hosts to 200 hosts.
• Mixture of research and DoD production nodes
• Included most major research universities labs

8
(No Transcript)
9
(No Transcript)
10
(No Transcript)
11

• ARPAnet Protocols
• IMP-IMP Protocol
• Reliably deliver a message (packet) of up to 8000
  bits to a specified destination host.
• Bit orientation to handle different word lengths
  of hosts.
• Using this IMP-IMP service, create a
• Host-Host Communication Service
• Network Control Protocol or NCP.
• Using this host-host service, can create
• User Application Services
• File transfer, remote terminal login, and email

12

• ARPAnet Protocols
• Network Control Protocol (NCP) -- why?
• IMPS only connect computers, but it is really
  programs that talk to each other.
• NCP can establish multiple logical data streams
  (virtual circuits or connections) between a
  pair of hosts.
• NCP opens and closes these connections.
• NCP provides flow control on each connection, to
  prevent overruning receiver application.
• NCP establishes an appropriate byte size on each
  connection.

13

• ARPAnet Protocol Stack

FTP File Transfer
User Protocols
NETRJS Remote Job Entry
Telnet Virtual TTY Terminal
SMTP Email
Host-Host Protocol (NCP)
IMP-host Protocol
14

• Legacy of ARPAnet Period
• o Demonstrated packet switching technology
• o Many new ideas...
• -- Protocol Layering
• -- Packet voice
• -- Email
• -- Link state routing protocols
• -- Connections (virtual circuits) o
  Reliable, flow-controlled requires state setup.
• -- Datagrams Pouzin, lt1974 o Single packet
  sent with minimum overhead o Unreliable
• -- Presentation Layer Postel White _at_ SRI
• -- RPC (Remote Procedure Call)

15

• Internet Research Period 1974-1983
• By 1974, other packet-switching technologies had
  been developed or were planned
• Local Area Networks (LANs) Ethernet
• Packet Radio network (packet switchradio now
  called ad hoc wireless)
• Packet-satellite network (SATNET)
• DoD's problem how to hook them together??
• Solution a network of networks -- an
  INTERNETWORK.

16

• Internet A Network of (Sub-) Networks

INTERNET
Host
Host
packet
subnet
R
subnet
R
R
R
R
subnet
Router gateway
Hosts
17
(No Transcript)
18

• Internet Research Period 1974-1983
• Seminal paper "A Protocol for Packet Network
  Interconnection", Vinton G. Cerf and Robert E.
  Kahn, IEEE Trans Comm., com-22, no. 5, May
  1974.
• ARPA funded Internet RD Program 1976 198x
• Developed Internet protocol suite ("TCP/IP")
• Application layer TCP/IP versions of ARPAnet
  application protocols Telnet, FTP, SMTP
• ARPAnet cutover to new Internet protocols
  TCP/IPJanuary 1, 1983 (Orchestrated by Jon
  Postel, ISI)

19

• Foundation Internet Technical Documents
• IP RFC 791
• ICMP RFC 792
• TCP RFC 793
• Telnet RFC 764
• FTP RFC 765
• SMTP RFC 788
• Name Server IEN 116
• Assigned Numbers RFC 790
• Although many contributed, these all bear Jon
  Postels name.

20
(No Transcript)
21

• 1984 - 1990 Academic Internet
• O NSF began to fund Internet growth through
  NSFnet and CSNET, bringing the Internet to US
  higher education.
• o Higher line speeds 56 Kbps -gt T1 (1.5Mbps)
  lines
• o Exponential growth gt repeated growth crises

22
(No Transcript)
23

• Important New Protocol Technology
• Domain Name System Map host names gt IP
  addresses
• e.g., "slime.usc.edu" -gt 32-bit IP address
• Later became the basis for URLs and the Web.
• Two-level routing hierarchy divide into domains
• IGPs (Interior Gateway Protocols) route within
  domains think area codes, plus
• BGP (Border Gateway Protocol) to route among
  domains
• Solution to congestion collapse Van Jacobson
• Network management SNMP

24

• 1984-1990 Who Ran the Internet?
• O US government agencies NSF, ARPA, DOE, NASA,
  had an important influence. (DADDY WAS STILL
  PAYING...)
• o Technical steering committee a junta of
  Internet researchers, the Internet Activities
  Board (IAB).
• -- Set up by ARPA (Vint Cerf) 1980 to advise on
  research program
• -- Until 1993, approved all Internet protocol
  standards.
• -- Created a set of Task Forces led by IAB
  members (1983)
• -- ONE of these task forces grew into the
  Internet Engineering Task Force (IETF), which
  today sets the Internet protocol standards.
• 3 IETF meetings per year, 2500 people, 100
  working groups.

25

• Who Runs the Internet Today?
• At no time has the entire Internet been run by
  one organization, nor is this likely to ever
  happen.
• The Internet is not brought to you by the phone
  company or by a government or by the UN or by
  Microsoft.
• There is no precedent for a global infrastructure
  built of thousands of separately
  controlled/administered pieces.
• It was designed to work that way, more or less
  but its still a minor miracle.

26

• 1990 - Now Commercial Internet
• Government funding replaced by commercial
  enterprises -- ISPs
• HTTP and the WWW were invented (1993) and
  rapidly became the dominant application.
• PCs became the dominant hosts
• Exponential growth accelerated!
• Huge increase in speed and capacity.
• The Internet became part of society.

27

• 1990 - Now Stresses on the Protocols
• o Growth crisis 32 bit IP addresses ran low
• gt developed IPv6 with 128-bit host addresses
• o Routing problems
• scaling, stability
• o New protocols and algorithms were developed--
  Mobile IP-- Security-- QoS
• -- Traffic Engineering

28

• The Internet Architecture
• Outline
• Internet Protocol Stack
• "Architecture" -- Fundamental design principles
• -- Guide specific engineering decisions,
  e.g., protocol design
• -- Want them simple and general gt elegant

29

• Internet Protocol Stack
• gt Application layer -- same as on ARPAnet
• gt Internetwork Layer IP (Internet Protocol)
• -- Controls end-to-end forwarding of packets by
  routers.
• -- Uses globally-unique IP addresses
• -- Best effort service packets may be lost,
  duplicated, reordered.
• gt Transport Layer
• -- Enhances host-host service e.g., TCP
  provides reliable full-duplex byte streams.
• gt Link layer particular subnet protocol.

30

• Internet ("TCP/IP") Protocol Stack

Application Layer
Telnet FTP SMTP HTTP POP3

...
UDP
TCP
Transport Layer
Internetwork Layer
IP
Link Layer
...
Ethernet
PPP
SONET
BBN1822
ATM
IP protocol gt plays a unique role as the
universal Internet protocol -- common bearer
service. gt Very simple, so routers can be
very fast gt Minimal service assumptions, so
Internet can adapt to new applications and new
technologies.
31

• Internet Architecture Deep Assumptions
• o Packet switching
• -- Unit of data is a packet
• -- Packets are statistically muxed Because
  computers dont hum to each other, they CHATTER!
• I.E., computer communication is fundamentally
  bursty.
• o Strict protocol layering
• -- Successive layers of functional abstraction
• -- Headers added/removed in strict LOFO order
  --
• Stackmodel.
• o Goal is universal interconnectivity

32

• Architectural Requirements
• o Generality
• -- Support ANY set of diverse applications, --
  Either datagrams (think video) or VCs (think
  web).
• o Heterogeneity
• -- Interconnect ANY set of network technologies
• o Robustness
• -- More important than efficiency
• o Extensibility
• -- More important than efficiency
• o Scalability
• (A later discovery. How many ARPAnets could
  the world support? A few hundred, maybe ?)

33

• The End-to-End Principle
• Foundation of the Internet architecture
• Dumb network, smart end systems.
• (Exact opposite of telephone network!)
• o Dumb networks only least common
  service -- Datagram service no connection state
  in routers
• -- Best effort all packets treated equally.
• o Smart hosts
• -- Maintain state to enhance service for
  applications.
• -- Fate-sharing-- If a host crashes and loses
  comm state,
• applications that are communicating share this
  fate.

34

• Fundamental Differences
• o Philosphical gap between Computer Scientists
  and telecommunication engineers
• gt Engineers "The Internet is under-engineered
  -- it is not optimal or completely predictable
  or controllable.
• (And we like complexity.)"
• gt Internet Researchers "Optimal is NOT the
  point.
• The future adaptability of the Internet to new
  technologies and to new services require that we
  NOT over-engineer the Internet! Uncertainty and
  suboptimality live with it! (And we like
  generality and simplicity.)"

35

• Conclusions
• The Internet survived and prospered ...
• o It survived a series of real (and imaginary)
  threats over the years.
• -- X.25
• -- FAX (opposing email)
• -- Many competitors from industry government
  -- e.g., XNS, DECNET, BITNET, MFENET,
• -- Unimaginative government bureacrats
• -- ISO Open Systems Interconnect (OSI)
• CLNP, TP4, FTAM,
• -- ATM
• o Exponential growth is a mighty engine, and
  vendors and ISPs have been running to keep up.

36

• Why did the Internet Succeed?
• 1. Leadership from a handful of visionaries
• Larry Roberts (ARPA) Vint Cerf (ARPA) Bob
  Kahn (ARPA) Steve Wolff (NSF)
• 2. ARPA-funded BSD Unix development used TCP/IP
• -- Publicly funded OS --gt many vendor products
• -- A generation of students used it and then
  graduated
• 3. Triumph of the startups
• -- Small companies were agile and creative

37

• Why ... (cont'd)
• 4. Military concern for "robustness" led to the
  easily-
• deployed IP protocol. o Plug-and-play --
  populist model fueled growth
• 5. Silicon- and Fiber-revolutions were perfectly
  timed.
• 6. The Web popularized the Internet.
• 7. Universal interconnection really is a powerful
  vision.
• 8. The Internet architecture provided the
  generality and extensibility to survive the past
  and the present.

38

• The Future is fuzzy.
• o Economic, business, and political forces are
  now central The Internet railroads in 1882
• Increasingly, large companies ruleo Will ISP
  mergers lead to monopoly?o Will the telephone
  companies rule after all?o Will Email survive
  spam?
• o Will spam, viruses, hostile attacks
  fatally threaten the grand vision of universal
  interconnection?
• o Will the Internet fragment into disconnected
  pieces?
• Incompatible pieces?
• o Will censorship take over the Internet?