 |
|
DAILY NEWS AND INFORMATION
FOR THE GLOBAL GRID COMMUNITY /
|
Special Features:
INTERNET PIONEERS HONORED BY
UNIVERSITY COLLEGE LONDON
By Christopher Lazou
University College, London (UCL) has been a European pioneer of the
Internet
for 30 years. As part of the 30th birthday celebration of the Internet at UCL,
Robert Kahn and Vinton Cerf, the USA Internet Pioneers were conferred the
Honorary Fellowship of UCL by the Provost. During the oration the audience
listened to a long catalogue of honorary Doctorates and awards both luminaries
received from other August institutions in their long and still active
careers. (Vinton Cerf wore a suave tunic, a cross between a monk's tunic and a
hippy outfit, which could only have originated in the West coast of America).
After the ceremony Mark Handley, professor of networks systems and protocol
expert at UCL, gave the inaugural lecture titled: "The Internet: the last 30
years and the next 30 years." Below is a synopsis of Vinton's 1988 keynote
speech to the Cray User Group (more detail of the past) and Mark's ambitious
lecture.
To reminisce, TCP/IP stands for Transmission Control Protocol/Inter-network
Protocol. The origins of TCP go back 35 years to the origin of the
ARPANET.
In 1968, ARPA (the Advanced Research Projects Agency); the D for defense,
was
added later to transform it into DARPA, issued an RFP for a new "packet
switch" engine. At the time ARPA was supporting about 30 projects at various
universities and other research establishments. The demand for the latest
computing equipment stretched even the ARPA budget. In addition, there was a
significant software development overlap at each site because of lack of
standards (UNIX was not invented yet).
To reduce costs, ARPA decided to build a communication network allowing its
research groups to share software and hardware resources. The then giant AT&T
advised ARPA that it needed some 450 point-to-point circuits to form a fully
connected network. At that time high speed meant 56Kilobits/second. The costs
were horrendous and even ARPA flinched on that. The alternative was a circuit
switch network, but this had a drawback, setting up circuits took many seconds
forcing machines to idle while waiting to send or receive data.
With the advent of the minicomputer and the concept of packet switching a
few
circuits (say 40) could provide connectivity among 30 sites with few "hops"
among a set of packet switches. As minicomputers operated with speeds
comparable to the mainframes, these switches did not need to set up circuits
before data could be sent. Each packet identified its destination and was
switched individually allowing all the host computers to automatically share
the high-speed telephone lines.
Although packet switching was initially greeted with scepticism in the
traditional circuit-switching world, a 20 node ARPANET was first demonstrated
at the International Conference on Computer Communications (ICCC), held at the
Washington Hilton Hotel, in 1972. In that demonstration, a protocol known as
NCP (Network Control Protocol) was used to support host-host communications.
Telnet and FTP lay on top NCP and supported remote terminal links and file
transfer services.
In 1973, Robert Kahn, one of the principal architects and a major force in
the
implementation of the ARPANET, left BBN and joined DARPA with the aim to
extend packet switching and by mid-1973 dreamed up packet radio and packet
satellite. In the same year Bob Metcalfe joined XEROX PARC and build ALOHANET
on a coaxial cable, which eventually became the Ethernet. Thus by mid-1973,
three new communications concepts came into being: packet radio, packet
satellite and Ethernet. About this time, Vinton Cerf, left UCLA and joined
Stanford where Bob Kahn introduced him to the problem of supporting reliable
communication across contention-oriented systems. The NCP fabric was not up to
this task therefore the new nets required a new protocol to support them.
During the summer of 1973, Vinton Cerf, Bob Kahn and Bob Metcalfe, met and
discussed the requirements the new protocol would have to meet. By September
1973, a joint paper by Bob Kahn and Vinton Cerf was presented at an
International Network Working Group meeting at the University of Sussex,
England. A more complete version of the design was published in another
Cerf/Kahn paper in May 1974 ["A Protocol for Packet Network Interconnections,"
IEEE Transactions on Communications, May 1974, Vol. Com. 22, No. 5].
In mid-1974, Ray Tomlison built a TCP at BBN, discovered a retransmission
problem and suggested a "three-way handshake" design modification to overcome
it. This idea was incorporated into the design and a full specification was
issued in December 1974. Implementations at BBN, Stanford and University
College, London followed in 1975. Under the stewardship of Peter Kirstein, UCL
connected via land link to Norway and from there via satellite to the ARPANET
in July 1973.
More tests and more amendments to cope with unearthed problems culminated
to
the ARPANET community converting to TCP/IP on January 1, 1983. TCP/IP spread
like bush fire. By 1988, some 150 vendors offered various products related to
the TCP/IP protocol suite. Once again the fabric -- this time the domain
address space was stretched -- and TCP congestion control had to be
introduced. By 1989 the NSFnet and BGP were up and running providing support
for policy.
After Vinton's keynote in 1988, I wrote an editorial in the Cray User
Newsletter, praising the achievements to-date and continued with the
following, "It is becoming painfully obvious that what is currently in place,
is not distributed computing, but the rudiments of things to come. If you
consider any other utility services from tap water to electricity to motorcar
facilities would you contemplate life at such primitive level?... it could be
unthinkable to consider the plumbing under the road when one takes a bath.
Only when computing services are on 'tap' we will feel that distributed
computing, has come of age."
In short, there was a lot yet to be done.
By 1990, the applications floodgates began to burst without warning
surprising
the experts. Email, telnet and ftp, were followed by Audio/video conferencing
and the World Wide Web by 1993, with browsing, hyperlink and so on. By year
2000 we had peer-to-peer file sharing which now accounts for 50 percent of the
traffic. In the last 7 years the people on the Internet rose from about 30
million to over 700 million, this amounts to over 10 percent of the world
population connected via the Internet. Thus, a lot has happened since 1988
including the Grid, but most of it unforeseen.
The future trend is for the Internet to become bigger and faster. It
approximately doubles every 16-18 months (Moore's Law growth) and 40Gbit/s
links currently deployed are capable to handle 2 million voice calls per
second. At this growth rate everyone in the UK could be continuously connected
for telephone and quality audio within 6 years.
There have been notable failures too. Examples include IP multicast -- one
to
many services, the introduction of IPv6 with its bigger address space. Quality
and protected services also are lacking. These may eventually succeed but it
could take many years. In the future many services can become accessible via a
mobile Internet symbiosis. News, event listings, train times, Google searches,
map-quest (multi-map), location information and the ultimate Holy Grail, the
ability of subtitling the real world.
The downside of the Internet is that it is general purpose and doesn't do
anything very well. As it gets faster, alternative special purpose nets with
quality and security controls are not build because they are not cost
effective. So we are stuck with viruses and Spam for the foreseeable
future.
The Internet has several immediate problems. It is running out of
addresses.
The current IPv4 uses 32bit addresses and even these are not allocated
efficiently. For example MIT has more addresses than the whole of China. IPv6
solves this problem but convincing people to switch to it, is nigh impossible.
Congestion control has insufficient dynamic range, routing policy from BGP, a
critical piece of Internet infrastructure, is also putting enormous
operational strain as nobody knows what policies are in use. But the biggest
problem is managing change of the infrastructure fabric.
In the next 30 years one can foresee the Ossification of the Internet, as
it
is already hard to change at the core. People would suffer from information
overload. Email, instant messenger, the Web, TV, Radio, DVD and so on would
ensure this overload. What is urgently needed is a way to filter out
irrelevant information, but unfortunately the net does not have embedded
knowledge. The other problem is conflict of generality and predictability.
Information is power. Connecting people in a decentralised way makes
censorship harder and brings the people of the world closer. It is not proper
to think of the Internet (or networks) as connecting computers, but rather
that they connect people who use computers throughout the world.
Having listened to Mark, I could not help recalling the quote by Alan Kay:
"It is easier to invent the future than to predict it."
|