The optimistic belief the Internet could make money
The technology became compelling
1.3 Why is there such a mess in Telecommunications?
All the players forgot the lessons of telephony networks
All capital intensive improvements need a rollout plan
Identify a customer that will pay for the improvements and
service
Historically this was business
Make the new service compelling
The money pouring into dot.com business stocks seduced
executives into believing that there was a new economics for networks.
If the new Internet Serviced Providers (ISPs) had a rollout
plan, it was ineffective.
There were too many crooks in the executive suite
2 Some History - It always takes longer than you expect!
Comment: Ever since the beginning of the Industrial Revolution,
there has been a continual belief that the The world has
never changed as fast as it is now!
The first telecommunication network was a data network - the telegraph
The customer was the Railroads who needed it for safety and control
A digital network with binary signaling and control
Email finally killed all forms of the telegraph network
In 1876, Alexander Graham Bell and Elisha Gray invented the
telephone. Gray lost his patent fight with Bell, setting the stage for
Bell to dominate the telephone industry
AT&T was so worried about network congestion that they
asked early residential customers to use the telephone only for
important calls
It took over 100 years until it was possible to telephone from
Canada/USA to Europe without an operator
2.2 Critical innovations for Telecommunication Networks
Automatic Switching
Step-by-step - Strowger (1892) - 16 years.
Crossbar switching - 1938 (Bell AT&T)
Computer controlled crossbar switch, 1965 (#1-ESS, Bell AT&T)
Fully Digital Switches: Nortel, 1972
A/D (Analogue to Digital) converters on every line card in the CO
switch
This invention and commitment allowed for an
end-to-end digital telephone network (well almost end-to-end)
IIASA (International Institute of Applied Systems Analysis),
Laxenburg, Austria - installed a new step-by-step in 1977.
The last manual Central Office (CO) switchboard, in Bryant Pond, Maine,
ceased operation in 1983.
Network Control
End/End Line pulses and trunk seizures (copper connections)
Network control using multi-tones (similar to "touchtone")
The separation of signaling and control from the transport of
voice.
This allowed advanced phone features such as 800 or "freephone"
service
198? - Advanced Intelligent Network (AIN) became the world's
largest packet switched network (SS7)
Bandwidth and Signal Quality
Satellite for long distance (Early Bird, 1964)
Use of fibre-optics in the core network and intercontinental traffic
Increasing fibre-optic penetration of the access network
Comment: There has been a continual improvement in quality,
capabilities, and features of the PSTN, but old breakthroughs such as
satellite channels are now considered poor quality compared to current
fibre-optic channels
Arpanet was justified by the US Dept. of Defense as a means of
sharing expensive and rare CPU cycles and software
In 1973 ARPA study showed 75% of Arpanet traffic was email.
The network was only 4 years old and this was totally unexpected.
The Internet has evolved from the first Arpanet node/host at UCLA
in 1969 to the January 2003, estimate of 171,638,297 host computers.
The original Arpanet was restricted to US Defense Contractors
(universities and industry) and created email envy in the computer
science community.
To satisfy the email envy, in 1981, the US
National Science Foundation (NSF) sponsored CSNET (Computer Science
NETwork) and then several versions of (1986, 1988, 1991, 1995)
NSFnet.
NSF allowed commercial use only from 1991/1995. By 1995, when
NSFnet was dissolved, it was the IntereXchange Point (IXP) for over 90
countries (April 1995).
1957: The Internet really began with the creation of ARPA
(Advanced Research Projects Agency) in response to
the launching of Sputnik.
1969: ARPA asked by US Department of Defense (DOD) to create a
network that would
Survive a nuclear attack
Allow DOD researchers at universities and contractors to collaborate
UCLA, SRI, UCSB, and Utah are the first four nodes
1970: Norm Abramson creates ALOHAnet in Hawaii - a random access
packet radio network
1973: Bob Metcalfe's PhD thesis at Harvard was the invention of Ethernet.
He then moved to Xerox PARC, and and built the first prototype, the
Alto Aloha System. Xerox did not want to commercialise Ethernet so
Metcalfe started 3Com a few years later.
1973/1975: Vint Cerf describes TCP to Bob Kahn and they eventually
refine it to TCP/IP. It took 10 more years (1983) before the
original NCP (Network Control Protocol) of Arpanet was replaced by TCP/IP.
1974/75: Arpanet includes satellite connections to Hawaii and UK
1981: CSNET (Computer Science NETwork) built with seed money by
NSF to get around use restrictions on Arpanet. CSNET was originally
one machine at BBN in Cambridge, Mass. BBN was the original prime
contractor for the Arpanet nodes.
1982: TCP/IP finally becomes the DOD network standard
"internet" protocol.
1986: NSFnet, established with 56kbps lines, connects several
super computer centers. NSF also supports local university networks that
are interconnected through NSFnet.
1988: NSFnet upgraded to T1 lines.
1990: Arpanet retired.
1991: After NSF removed commercial use restrictions on use of
NSFnet, the Commercial Internet eXchange (CIX)
Association formed by CERFnet, PSInet, and AlterNet (UUNET), IXPs become
the key method for interconnecting networks.
1991: NSFnet upgraded to T3 lines.
1995: NSFnet dissolved. This was the critical point in the
Internet evolution and set the stage for the current internet
architecture. At the time NSFnet was dissolved, there
were more than 90 countries connected to it. It was truly the world's
network hub/switch.
Ethernet has become the dominant access because the absolute
limits of performance have been finessed over the years.
It is an open technology
Current Ethernets are usually 10BaseT, 100BaseT, ... based,
using twisted pair connections and are full duplex - a marked contrast
from Metcalfe's original coax system.
7.2 An evolving? Internet Architecture for Real Time Traffic
Local Access (Ethernet???)
Most academicsolutions use tokens
Ethernet switches can control delay at access not up the network
Metropolitan Networks (Gigabit Ethernet???)
Continental and Intercontinental Networks (Architecture ???)
Routing and Bandwidth Allocation
Effective Bandwidth (Kelly) to ensure appropriate delay
Effective Bandwidth is not applicable if the traffic really
is fractal or heavy tailed
Traffic Categorisation must not compromise delay
Key: Improve the underlying performance of the network
1999 - Backbones had delays of seconds and losses of 25%+
2003 - AT&T claims losses of 0% and delays of 10ms in USA and
250ms to Japan
7.3 IETF Working Groups Directly Concerned With Real Time Traffic
Most IETF efforts require too much processing connection tracking
to be viable for very high volume/speed traffic. This appears to have
led to the demise of QoS Routing and RSVP working groups
Order preserving routing is preferred for flows (all TCP
connections, RTP, and real time UDP connection.)
Problems:
Route Flap: a characteristic of BGP4, OSPF - fundamentally due
to routing update synchronisation and the lack of any calculation to
estimate the change in network delay topology by a change in routing -
at this point, one lives with it as a fact of life.
Load balancing: intentionally spreading packets amongst equally
good (bad?) routes
RED: Rapid Early Detection - random discarding of packets by a
router during periods of congestion. Intent is to prevent all the TCP
connections from retransmitting at the same time and causing a
"blizzard" of additional packets. UDP ignores this action.
For IPv4, special cases are bumped out of the fast-track
processing for slower, special treatment. It remains to be seen whether
the IPv6 header design is sufficient to allow unprejudiced use
of options.
7.5 Requirements for Real Time Flows On The Internet
Sequencing: reordering in real time at reception, and
compensation for loss without retransmission;
Intra-Media synchronization, for proper restitution, including
silence periods (not transmitted);
Packet Voice was a goal in the early years of the Arpanet -
Voice quality in these early Arpanet experiments was dismal.
Low speed - 1200bps to 56kb
Multicasting, for real time multimedia conferencing, was proposed
in 1985 by Steve Deering as his PhD thesis at Stanford.
Multimedia distribution via multicasting is still very limited. The
MBONE (Multicast Backbone) is the current experimental subnetwork for
multicast.
"Real Audio on the Internet" is an example of streaming real time
traffic - a fixed delay is added to smooth out jitter.
IP proponents of conferencing over the Internet claim it works with
acceptable quality (Huitema, Crowcroft). Advocates tend not to be too
discriminating.
IP telephony exists and current pricing reduces complaints.
IPv6 was designed to help improve real time traffic.
RSVP was supposed to be the protocol of choice to handle QoS, but it is
now viewed as being unnecessarily complex, which has made it unappealing
for the community at large and even its working group has disbanded.
Reserves resources for multicast conferences
Receiver based - the receiver builds a multicast tree from itself
to all the senders in the conference, reserving resources as it is
built. Since there is, usually, only one sender active at time,
reservations for the same resource may be shared.
Basic assumption is that receivers will have widely differing
capabilities and only the receivers know what they are. This precludes
sender resource reservation
Both receivers and senders in a conference are undisciplined
coming and going at will.
The key idea is to annotate IP packets with a small, easily read field,
and have the Internet router/switches handle them differently. This
appears to be a proposal that could, eventually and radically, change
the basic communication infrastructure of the Internet. The
diffserv working group is developing the rules that compliant IP
routers/switches should/must follow:
Supporting Technologies
IP switching - Tag, Label (ATM?)
QoS based routing
Separation of Route calculation and Packet forwarding
Question: ATM and IP: When does a long lasting IP packet/flow
based route become equivalent to a virtual circuit - or does it ever?
8 A Bottom Line - Internet Architecture for the New Infrastructure?
IPv6, or something resembling it, will replace IPv4. IPv4 is not
sufficient for universal access
Variability in all performance measures is reality
Internet architecture resembles telephone architecture at the turn
of the century, with the world being the village, and many outlying
"farms" still not having service
The "near term" communications infrastructure will be a melange
of current switching techniques
The current Internet will be "current", for only a few days
9 A Potential Architecture for the New Telecommunication
Infrastructure
9.1 IP based packet network communications infrastructure
New IP-based networks for integrated services are being proposed for
small, medium and large size corporations.
Will they work? As long as the end-to-end network is small enough ...
Will it be cost effective? Switching packets may be cheaper, on a small
scale. On the WAN side, the infrastructures are similar.