Gigabit Networking - Seminarsonly

deadpannectarineΔίκτυα και Επικοινωνίες

26 Οκτ 2013 (πριν από 3 χρόνια και 7 μήνες)

89 εμφανίσεις



Synopsis ………………………………………………………………………. 1

1. Introduction to Gigabit Networking ………………………………………….. 3

2.Gigabit Tec
hnologies and Concepts ………………………………………….. 5

2.1 Fiber Optics …………………………………………………………………6

2.2 Cell Networking ……………………………………………………………7

2.3 Packet Networking…………………………………………………………..11

3. Trends and issues

in Gigabit Networking………………………………………12

3.1 Challenges in Gigabit Networks……………………………………………13

3.2 Gigabit Testbeds…………………………………………………………….14

4. Gigabit Applications……………………………………………………………15

5. Gigabit Ethernet………………………………………………………………...

6. Technology Elements Required for Gigabit Networking……………………… 24

7. Conclusion…………………………………………………………………… 30

8. Bibliography ………………………………………………………………… 32


Several industry trends are leading enterprise users to examine the need for gigabit networks.
Each company situation will be different, requiring a specific set of migration steps to handle
growing network traffic and changing traffic pa
tterns. Key industry trends creating the
evolution toward gigabit networking include the emergence of the intranet network model,
higher bandwidths required by network users, increases in processor power, new
applications, and changing traffic patterns.

ithout question, today's most significant drivers within the

enterprise network are Internet,
intranet, and extranet technologies. This is reflected in the explosive use of Web technologies
that are fundamentally transforming the manner in which business i
s conducted. In addition,
a higher number of network users depend on traditional applications (such as

file transfer, e
mail, and network backup) to conduct business, creating a steady growth of network traffic.
The result is a geometric growth in traffic
and a permanent change in the nature of enterprise
networks, as well as an increase in

the commercial assimilation of both protocols and
application styles that began life on the Internet.

One of the most significant changes that has occurred is

the unpre
dictable network traffic
patterns that result from

the combination of intranet traffic, fewer centralized campus server
locations, and the increasing use of multicast applications. The old 80/20 rule, which stated
that only 20

percent of network traffic we
nt over the backbone, has been scrapped. The ease
with which internal Web browsing now enables users to locate and access information
anywhere on the corporate intranet means that traffic patterns are dictated by where the
servers with the most valuable pa
ges are and not by the physical workgroup configurations
with which they happen to be grouped. Thus the vast majority of

traffic will traverse the
backbone, and any
any traffic will become the rule. Simplified access has also sent overall
usage of the n
etwork skyrocketing, as users point and click their way through the corporate
portfolio of Web
based resources.

Without question, intranets are a key design center for the implementation of sophisticated
multimedia application styles that are becoming in
creasingly complemented by the use of
infochannel push technology. Complement this trend by the doubling of uniprocessor power
and the quadrupling of multiprocessor power every eighteen months, and you have the basis
for nearly geometric growth in network
performance requirements that are being generated
on a

year basis.

The long
term implication is that users will increasingly demand solutions for

whose specific requirements include tens if not hundreds of gigabits per second

total network capacity, any
any communication, smooth migration of scalable
performance that can be incrementally implemented anywhere in the network, and

compatibility with the infrastructure of existing enterprise networks.

Growing numbers of
users on the network, more current applications, faster desktop
computers, and faster network servers create a demand for higher
performance LAN segment
capacity and faster response times. Bandwidth enhancement beyond Fast Ethernet is needed
to provide smo
oth network operation in the face of emerging bandwidth requirements.
Improvements in network
layer performance to hundreds of thousands of packets per second
or millions of

packets per second is required to meet the challenge of

ayer throughput and changing traffic patterns.

1. Introduction to Gigabit Networking

Computers and their attachments (like networks and disks) are getting faster everyday. The
current CPU speeds of processor like DEC Alpha and Pentium are well ove
r 100MHz
allowing them to perform billion instructions per second (BIPS). This speed is comparable to
supercomputer's speed five years ago. With the growing speed of computers the applications
which run on them are now ranging from interactive graphics, vo
ice recognition, video
conferencing, real time animations etc. All these new applications will use networks to carry
more data.

Network bandwidth is also increasing concurrently with the CPU speeds. When in 1980s
10Mbs Ethernet was considered fast, we now

have 100 Mbs Ethernet. The bandwidth is
approaching the speed on 1 billion bits per second (1 Gbs), much due to the research in the
field of fiber optic signalling.

The three main fields data communications, computing and telecommunications are
g a period of transition. The field of computing is rapidly advancing with processor
speed doubling ever year. The latest
(Redundant Arrays of Inexpensive Disks) has
given rise to file
systems with gigabit

The field of data communications
which facilitates the exchange of data between computing
systems has to keep up with the pace of the growing computing technologies. In the past the
data communications provided services like the
. Now applications like virtual reality,
video confere
ncing, video on demand services are present.

For a century the telecommunication industry has been carrying voice traffic. This scenario is
changing with telephone networks carrying more data each year. The data being carried by
telephone network is growi
ng at 20% per year compared to voice traffic which is growing
only at 3% per year. Soon the data traffic will overtake the voice traffic. All this, has made
the telecommunication industry more interested in carrying data in their networks.

So the three
communities are now converging with common interests of carrying more data
at higher speeds. This has led to some joint activities. The most notable of these activities is
that which has led to the setting up of gigabit testbeds in United States. Other joi
nt activites
are the standardization of ATM (Asynchronous Transfer Mode), a suite of communication
protocol to support integrated voice, video and data networks. Some organizations which are
doing research in gigabit networking are
National Coordination Office for HPCC

Performance Computing and Communications)
, IEEE Communications Society Technical
Committee on Gigabit Networking.

When the gigabit networking was in its horizon, many researchers felt that

the current
knowledge about networking would not apply to gigabit networks which are considerably
faster than existing networks. Now, after several years of research it has been found that
many of the strategies and techniques (like layering the protocol)

still work in gigabit
networks also.

There are many working Gigabit testbeds .In five to ten years Gigabit networks will become
a reality. It is now unclear whether there will be a single gigabit technology with a specific
standard protocol. But it looks

like that there will be many competing gigabit networking
technologies (like many LAN technology) and many protocols but eventually one of them
will become most popular (like IP).

2. Gigabit Technologies and Concepts

The development of high spee
d networks is closely linked with the advancements in fiber
optics. The advent of fiber optic signalling equipment capable of transmitting at several
gigabits per second over long distances and with low error rates through optical fiber showed
that gigabit

networks were feasible and has served as a goad to researchers. Other media like
radio and micro waves also have been explored and found to be capable of providing gigabit
bandwidth. The LuckyNet is a 2.4 Gbs (OC
48) testbed link build by AT&T. The Advanc
Communications Technology Satellite (ACTS) is an experimental satellite that was launched
by NASA in July 1993, which supports OC
12 links in one configuration.

Another important trend in gigabit networking is the increasing interest in a technology no
widely known as
cell networking, cell switching or cell
. In the following (sub) sections
we will give a brief overview of these concepts and technologies.

2.1 Fiber Optics

2.1.1 Fiber Optic Basics

Some part of the light gets reflected when passi
ng from of one medium to another, and the
rest of it gets refracted. Light has an interesting property that if the angle of incidence is
greater than a critical angle that all of the light is reflected. Fiber Optics uses this property of
the light in sendi
ng the signals.



The fiber has a thin strand of glass called
surrounded by a thicker outer layer

is sent at the appropriate angle inside the core and it travels through the core, any light
escaping the core will be reflected back into the core. The pulses of the light carry the bits.
One important thing here is that the bits dont travel faster (propa
gation delay is similar to that
of copper wire) than in copper wire, the higher bandwidth is got because the bits can be
packed more densely (1000 times more than that of copper). Theoretically the fiber has
bandwidth of 25 THz around each wavelengths of 0
.85, 1.3, 1.5 microns. (These give rise to
bands similar to those of radio waves). So the total capacity of a single fiber is 75 Terabits
per second!

In fiber optics just like copper wire there are some problems while signalling. There are three
major typ
es of dispersions: modal, chromatic and material. Repeaters and Amplifiers which
strengthen the signals are used to overcome these dispersions. Also
fiber (also
) is used to avoid dispersion. The
fiber still suffer from
dispersion problems.

Transmitter and receivers are generic terms for devices attached to a fiber to respectively
transmit and receive signals. These have two important varities: fixed and tunable. Fixed
ones are set to a particular wavelength and tu
nable ones can dynamically set the lightwave
frequency at which they transmit or receive. Fixed ones are simple, the tunable onces are
more complicated. One example of a tunable device is the Mach
Zehnder Inferometer. Here
one path of the light is made sli
ghtly longer than the other such that there are out of phase
and the difference is used for tuning to a particular wavelength

2.1.2 SONET and the SDH

SONET the telephony standard which stands for Synchronous Optical Network (called the
Sychronous Digitia
l Hierarchy, SDH in Europe). SONET was developed to support
multiplexing on links capable of data rates of hundreds of megabits or more. Its main goal
was to provide a single set of multiplexing standard for high
speed links. The standard
provides with var
ious rates called as
Synchoronous Transport Signal levels
(STS) or

Carrier level
. The rates range from 51.84 Mbs (STS
1) to 2.4 Gbs (OC
48). SONET
carries the data in frames. Each frame has overhead and payload part.

2.1.3 WDM Networks

contrast to the SONET, WDM networks uses the special properties of the optical fibers. In
WDM networks the bandwidth of the fiber is divided into multiple channels and hosts
communicate to each other on a particular channel. There are two major types of WD
. As the name suggests in
WDM networks the hosts
are directly connected to each other via a star
coupler. The
can be further
classified depending on whether they use fixed/tunable transmitters and receive
rs. Two
examples of
WDM networks are LAMBDANET a project of Bell
Communications Research and RAINBOW a project of IBM. The
WDM Networks
can be designed in many ways. The main goal is to build a high
connectivity graph so that
the number o
f hops between any two nodes is minimized. An example of
network is the TeraNet, which was built as the part of ACRON project at Columbia

2.2 Cell Networking.

2.2.1 Fundamentals

The ideology of cell networking is that all the data
should be transmitted in fixed size packets
. In networks usually the data is sent in
which vary in size. If the packets
vary in size it is difficult to guarantee bounded delays which are required for isochoronous


The waiting time is large when the small
packet of the isochoronous traffic (eg. voice traffic)
is waiting behind a large packet. In contrast in cell networking, since the large packet is
divided into small cells the waiting time is not large. In cell networking it possible to give
guarantees for


There are two ways of switching cells, one method is
and the other is
. In
switching the whole cell is received and then it is forwarded in
a appropriate link. In
switching the switch decides

on which link to forward the
cell by just examining a few bytes of the header in the cell. Asychronous Transfer Mode is
the brain
child of the cell networking technology. The telecommunication community has put
in a lot of effort to standardize this techn
ology. There is also a
ATM Forum
, which is a group
which is deciding on the main issues of the ATM before waiting for the official standard
come out. Currently ATM provides bandwidth from 45 Mbs to 622 Mbs. But in future, ATM
will also provide gigabits of

2.2.2 Cell Networking in LANs

Most of the Local area cell networking share the media. Since the media is shared the issues
of who gets the access of the media has to be taken care. The Local area cell networks solve

this problem by arbitratin
g the right to send a cell. Most the local area cell networks are ring
networks. Now we shall see some examples of local area cell networks.


Cambridge Backbone Ring (CBR) was developed as collaborative project between
University of Cambridge and Oliv
etti Research. It is a ring network in which the ring
is divided into
. Special five
bit patterns are inserted between the frames for
synchronization and to equalize timing. CBR allows slots to be grouped.

IEEE 802.6 (DQDB)

The Distributed Queue Du
al Bus is joint standard of Institute of Electrical and
Electronic Engineers (IEEE) and the American National Standards Institute (ANSI).
DQDB is the fore
runner of ATM, it has same cell format has the ATM. It uses a dual
bus in a slotted
ring network. Eac
h slot can hold one cell of 53 bytes (48 data + 5
header). The dual bus is advantageous because the network can reconfigure to form a
bus when one node fails. Unlike the CBR, DQDB does not allow cells to be grouped.


This is a prototype gigabit ce
ll LAN built by Hewlett
Packard Laboratory in Briston,
England. This differs from the CBR and DQDB in that it uses a large cells (256
bytes) because most of the LANs use large packets. It uses a logical folded bus
architecture, each node is attached twice,

one with the write bus and one on the read

bus. One nice feature of the HANGMAN is that if the node wants to send a packet
which is twice as long as the cell, then it can request for two consective slots and send
the packet.

Another issue in local area

cell networking is
. The
is ATM switch was
built by Digital Equipment Corporation's Systems Research Center. It is a small (16x16)
cross bar switch. The traffic is divided into continuous bit
rate and variable bit
rate virtual
It can support data speeds as high as gigabit per second. The continuous bit

traffic must preallocate the bandwidth with the switch. To prevent loss in variable rate
circuits, the switch uses a combination of input buffering and hop
hop flow contr

technique. It has a auto
learning feature, by which the switch learns the presence of other
AN2 switches and this learning can be used for routing packets.

2.2.3 Wide Area Cell Networking

Switch design is key issue in wide area cell networking. The
switchs will have to switch
reliably very large bandwidths since it is operating at gigabit speeds. Usually these switches
are designed using parallel interconnection devices. A fundamental problem in switching is
happens when two cells c
ontent for a particular link. Another important
issue is the buffer management. Various buffering strategies like
internal buffering
are used. We shall briefly see some of the wide area cell
networking switches.



In these switches every input is connected to every output by a cross bar. These switches use
output buffering (cells are buffered at the output port). Multicasting can be easily provided in
bard switching. The crossbar switch usually have

very low blocking probablities.
Hence these switches are used as a standard for comparing other switches. Two examples of
crossbar switches are the
knockout switch
Guass switch

The Banyan
Batcher Switch

The main disadvantage of the crossbar switch
is that it uses O(n*n) circuitary for connecting
n output ports. This switch uses only O(n*log(n)*log(n)) switches. This switch assumes that
the n links have uncorrelated (independent) data links. It uses the batcher sorting element for
providing connecti
vity from any input node to any output node. The
examples of Banyan
Batcher switches.

2.3 Packet Networking

Packet networking has been around since the advent of networking field. More research have
been done on packet networking co
mpared to cell networking. At gigabit speeds since the
propagation delay of the medium is comparable to transmission time of the packet (eg 1.2

microseconds for transmitting 1518 bytes at 1 Gbs), problem due to serialization (large
waiting time of small
packets) are not present. Another problem the packet networks have to
overcome is that they have to support isochoronous traffic. It has been shown that adaptive
applications (applications which change their behaviour) can be used to overcome the
problem o
f providing isochoronous traffic. Already some high
speed LANs and WANs have
been built using the packet networking.

2.3.1 Local Area Packet Technologies


This is a gigabit LAN built by USC Information Sciences Institute. It is built based
chip developed by CalTech. The Mosaic chip uses a 4x4 or 8x8 mesh to
interconnect 0.5 gigabit links. The chip routes packets by having
counter and
counter. The difference between the source and the destination is used to move the
packet along th
e x and y coordinates. Mutlicasting is supported by having the
processor and each concentartor to duplicate the packets.

The CSMA/RN Ring

This is a ring network (a follow on of the FDDI), developed and studied in Old
Dominion University. Data from the ri
ng is put in a short delay buffer. If the ring is
not active (no data in it) then the controller can send the data immediately. Collisions
are avoided by placing the new data from the ring at the end of the to
packet in the delay buffer. Ano
ther good feature about CSMA/RN is that the packets
are removed at the receiver.

2.3.2 Wide Area Packet Technologies

The main problem of wide area packet networks are the router which have to route packets at
gigabit rates. Forwarding can be done using ve
ry small number of instructions (100
instructions, because for forwarding, the router just examines the header and makes some
consistency checks and routes the packet to the appropriate ports. Using multi
instructions, the CRC calculation can

be done while the packet is being copied itself. Also

various improved hashing and lookup table techniques have been developed which are used
in these routers. We shall see two high speed routers


A High Speed Packet Router

It is a 8 port ring
shaped bus switch developed by Cidon and Gopal of IBM. It uses source
routing, in which each adapter in the ring examines the packet and transmits the packet if the
source routing specify it to be transmitted. Multicasting is provided by adapters replicati
the packet. The packet is not removed after transmission, it circles round the ring once and
then removed.

The Bell Labs IP Router

A team at AT&T Bell Laboratories has developed a prototype gigabit IP router. It uses
multiple processors, the IP packet

header is stripped and passed onto one of the processor
which updates the header and sends it to the appropriate outbound processor. Its rate is just
over a gigabit, but it is precursor of the future gigabit IP router which use multiple processor
for rout

3. Trends and Issues in Gigabit Networking

There is voluminous amount of research in the area of gigabit networking currently. The
IEEE Communications Society Technical Committee on Gigabit Networking has been
conducting a yearly Workshops on Gigabit
Networking. In the following sections a brief
summary of current trends and issues in gigabit networking is presented.

3.1 Challenges in Gigabit Networks

The major challenges in networking research are to take advantage of the newly developed
techniques f
or building high
speed networks, and find ways to evolve it to meet new
applications needs, to keep pace with other computing technologies, and to encourage the
transistions of gigabit technologies into the wider community. To achieve these goals the
it Netowrking Workshop identified important problems in the following areas :

Performance evaluation

Higher speeds and new traffic mixes are causing a much needed re
examination of models
and algorithms for networking performance.

Switching technolo

Achieving higher speeds and new types of traffic are forcing the networking community to
develop innovative techniques for minimizing the cost of per packet processing in switches
and routers. A continuing challenge is finding a way for these techniqu
es to scale to switch
designs with more connections per switch and higher bandwidths per connection.

Network management and control

A combination of new types of traffic, larger bandwidths, and the long relative delays in
gigabit networks have made the p
roblems congestion control and finding routes for data
transfers substantially more difficult. Research is needed on how to best balance congestion
control between the network and end
systems and on methods to quickly find valid routes for
new data transfe


Gigabit networking technologies will have to interoperate both with each other and with
existing networking technologies. As a result, internetworking will be at least as important in
the future as it is now. While the basic ideas of I
P architecture apply to gigabit networks it is
also true that our internetworking technology needs to evolve to take advantage of the new
capabilities of gigabit networks.

Interfacing computers and application to networks

While it is now feasible to deli
ver data at gigabit rates to a computer's interface, we continue
to have great difficulty getting the data through the interface and computer's operating system
to the application quickly and at gigabit rates. Considerable work, probably in conjunction
h the operating system community, is needed if applications are to use gigabit networks to
their full potential.

Gigabit interfaces for PCs

Gigabit networking is no longer the domain of supercomputers and high
end workstations.
PCs will soon need gigabit

capabilities too and we need to encourage the development of
interfaces with low costs and low heat and power consumption.

end protocols

Better ways to develop end
end protocols that meet the needs of applications are needed.
Ideas like the
ability for applications to synthesize new protocols from functional
components need to be explored.

Shared media access technologies

The traditional thinking is that the high bandwidths and relatively long delays in gigabit
networks limit our choices of

local media access techniques but emering research suggests
that there may be a wide diversity of media access tecniques that work for gigabit networks
and these options should be explored.

Parallel channels and striping

It is often more cost effective
to send data in parallel over multiple links than send the data at
a higher bandwidth over a single link, a technique known as striping. While the idea of
striping is well
known, it is still inadequately understood.

Design and verification of protocols


tremendously frustrating problem in networking is our inability to design protocols of even
modest sophistication and prove that they work correctly. Some new ideas are being
developed in this area which combine design with formal verification, and given
that current
verfication technology is nearly 15 years behind the rest of the field, we need to encourage
new work in this area.

3.2 Lessons from Gigabit Testbeds

Through a collaboration of industry, academia, and goveernment, work on five US testbeds

begun in 1990 with funding from National Science Foundation (NSF), the Advanced
Research Projects Agency (ARPA), and industry. The testbeds are known as
Aurora, Blanca,
Casa, Nectar,

The following are the lessons learnt from these and other

Gigabit Testbed initiatives.

The existing gigabit networks testbeds have proved that gigabit networking
technology is feasible. So, the question is not "Whether we will have gigabit networks
in future?", the question is "When we will have gigabit network
s available?"

Initially researchers believed that the existing networking techniques will not be
applicable in high bandwidth. The testbeds have proved that most of the existing
knowledge can be applied to gigabit networking as well.

A few applications h
ave been implemented in the existing gigabit testbeds and have
shown the utility of such networks. Some Grand Challenge class scientific
applications (e.g quantum chemical reaction dynamic, global climate modeling,
chemical flow
sheeting, traveling salespe
rson problems) have solved problems that
were previously out of reach by using the network to combine geographically
distributed computing resources. When heterogeneous computer architecture were
involved, some applications achieved remarkable superlinear

Parallel computing techniques were used for programming the metacomputer created
by combining sytems across the gigabit networks. Distribution of the application
programs across multiple machines was typically based on the functional
on. Different tasks or phases of the program were parceled out to the
connected computers and these tasks exchanged information during the
computation as needed. Pipelining approaches were used to hide the latency of
networks communications.

it testbeds have also helped in indentifying the networking problems which
arise due to high bandwidth. Many of these challenges are mentioned in the previous

4. Gigabit Applications

Most of today's data network applications are not very sensitiv
e to delay and variations in
bandwidth. It does not matter very much if your files take a little longer to travel across the
net. But in telecommunications (telephone) industry the applications are delay sensitive.
Normally, when humans speak they pause in

between sentences and if the pause is longer the
other speaker speaks. If due to network delay if the pause is large then both the speaker may
speak at the same time leading to confusion. So delay should be bounded in telephones.
Other applications like X

Windows, remote login etc will be faster if gigabit networks are

Any applications which needs low response time or high bandwidth is suitable to be a gigabit
application. The recent advent of gigabit networks have given rise to many new applicatio
One of them is IVOD (Interactive Video on Demand). Here the consumers order which ever
program they want to see and the programs are sent from a centeral server to the consumers.

Since video applications use a large bandwidth and also different viewer
s may want to see
different programs at the same time, this application will benefit from gigabit networking.
Though compression methods like
may be used for compression, every once in a
while full screen data has to be sent. This can be done using gig
abit networking technology.

Highly computation intensive problems can be broken into smaller problems and given to
computers with high bandwidth networks connecting them for interchanging data. For
example in UCLA, researchers are experimenting with simul
ation studies of atmosphere and
ocean interactions. One supercomputer (CM
2) simulates the ocean and another simulates the
atmosphere and these interchange huge amounts of data and the interactions are studied.
Typically 5 to 10 Mbs of data is exchanged pe
r cycle, this will take a second in a 10 Mbs
Ethernet while it will take only 100ms in a gigabit networking environment.

Another class of applications are those which have real
time interactions with humans. A
typical example is video conferencing. Humans

are capable of absorbing large amounts of
visual data and are very sensitive to the quality of the visual data. Another class is the
applications which give the user the illusion of being somewhere else. There have been
interesting experime
nts done in NASA. They developed a system, by which the geologists
can interact with the surface of the Mars. Geologists study by interacting with the surface,
touching it (virtually), seeing the 3D scene from different angles etc. All these require large
amount bandwidth, and gigabit networking comes to help them out.

One of the main difference between the traditional data
communication applications and
those of interactive nature is that the later need timing requirements about spacing between
samples. R
ecently there has been some work in some innovative experiments with
. These applications change their behaviour dynamically and require only loose
performance guarantees from the network. An example of the adaptive application is the

conferencing system developed by Van Jacobson.
is like a telephone but uses the
computer and internet to connect two peoples. The
avoids isochoronous samples by
keeping a large buffer and timestamping all the data it receives. Traffic whic
h arrives earlier
are buffered appropriately and then played. The time of the delay is called the
So adaptive applications could have a major impact in network designing in future. The vat
also has demonstrated that slow networks like Intern
et can support real
time applications
with enough buffering.

5 Gigabit Ethernet Establishes a Foundation for

Gigabit Networking

5.1 Gigabit Ethernet Is a Natural



The growing importance of LANs today and the increasing complexity of desktop c
applications are fueling the

need for high
speed networks. The bandwidth provided by a 10
Mbps Ethernet connection may not be an adequate match for today's typical desktop
computing applications.

Numerous high
speed LAN technologies have been prop
osed to provide greater bandwidth
improved client/server response times. Foremost among them is Fast

Ethernet, or 100BaseT,
a technology designed to provide a

nondisruptive, smooth evolution from 10BaseT Ethernet

speed 100
Mbps performance. Given
the trend toward 100BaseT connections to the
desktop, the need for

even higher
speed connections at the server and backbone level is clear.

Gigabit Ethernet will be ideal for deployment as a backbone interconnect between
10/100BaseT switches and


a conn
ection to high
performance servers. A natural upgrade
path for future high
end desktop computers, Gigabit

Ethernet will require more bandwidth
than can be

provided by 100BaseT.

5.2 The Emergence of Gigabit Ethernet

The IEEE 802.3z Task Force has issued
a draft for Gigabit Ethernet. The draft allows for
both half

and full
duplex operation with a variety of physical interfaces. Thus, there will be
switched and shared topology implementations for Gigabit Ethernet (see Table 1). The
choice of topology will

depend on the network connection objectives. For example, switched
topologies provide the longest distance and high throughput. Shared topologies will provide
lower cost with shorter distance capabilities. Cisco is developing products that are targeted

switched topologies and that will be compatible with interfaces that operate in shared
environment (half
duplex mode).


Table 1: Gigabit

















Wiring closet



Low cost







Desktops (long

Low cost


Full duplex





Desktops (long

5.3 Gigabit Ethernet for Campus Intranet Applications

One application for Gig
abit Ethernet is the building backbone. For this application, Gigabit
Ethernet is deployed for backbone links in the building riser that connects a centrally located
switch with each wiring closet. Each wiring closet switch has a Gigabit Ethernet uplink.
ultimode or single
mode media is used to achieve the required distance. A

Gigabit Ethernet
switch is centrally located in the building data center with connection to servers, routers, and
Asynchronous Transfer Mode (ATM) switches as needed. The server conn
ections can use
copper or

distance fiber for lower cost. Routing and ATM services are provided as
needed for high
speed connection to

the wide
area network (WAN) or metropolitan
network (MAN).

A second application for Gigabit Ethernet is the cam
pus backbone. Here, Gigabit Ethernet
links are used to connect switches in each building with a central campus switch. Full
operation achieves maximum throughput and distance with fiber media. Either single
or multimode fiber can be used. A Gig
abit Ethernet switch is located in a central location
with connection to servers, routers, and ATM switches as needed. The server connections
within the campus data center can use copper or short
distance fiber for lower cost. Routing

and ATM services a
re provided as needed from the campus data center for high
connection to the WAN .

Figure 3: Building and Campus Applications of Gigabit Ethernet

Gigabit Ethernet is well suited for connecting high

performance servers to the network.
Servers are growing in

power and throughput. Processing power is doubling every


months. This growth, combined with the trend for

centralizing servers within large
enterprises, results in a requirement for very
bandwidth network connections. Today,
performance UNIX servers are able to flood three to four Fast Ethernet connec
simultaneously. As the processing power of these systems grows, they will require a faster
network. Gigabit Ethernet is ideally suited to

provide the high
speed network connection.

In specific applications such as animation, film postproduction, or i
mage processing that
require transfer of

larger files between desktops and servers, network performance is directly
proportional to business productivity. Gigabit Ethernet will provide a solution to current
network performance constraints. In the short ter
m, Gigabit Ethernet is expected to be
deployed for backbone and server connections. As desktop systems continue to grow in
power and as applications require more bandwidth and faster response time from the

network, Gigabit Ethernet will be deployed at the

desktop to meet these specific

5.4 Where Gigabit Ethernet Fits

Advantages offered by

Gigabit Ethernet include the following:

cost bandwidth

CoS based on the RSVP and the emerging IEEE 802.1Q/p standard, which provide
differentiated ser
vice levels

Leverage installed base of Ethernet, Fast Ethernet, and

LAN protocols

Leverage installed base of Ethernet knowledge for management, monitoring, and

Cisco Gigabit Ethernet products are designed for campus backbone, building ba
ckbone, and
server connection solutions. For customers who will be deploying video and

voice over IP,
Gigabit Ethernet, along with Fast Ethernet, will provide a low
latency IP infrastructure to
deliver these services throughout the enterprise campus. Until

Gigabit Ethernet products are
available, Fast EtherChannel can be

deployed to scale bandwidth beyond Fast Ethernet

switch, router, and server connections.

5.4 Cisco is Uniquely Qualified to Meet the




Cisco has been a pion
eer and is now the industry leader in

routing and switching solutions.
Cisco recognized that the convergence of routing and switching technologies is

required for
the delivery of high
performance, scalable campus networks. Today, Cisco is fusing router

switch technology to develop solutions that address the challenges of

gigabit networking
while providing a smooth migration path to gigabit network performance. This class of
product is

generally known as gigabit multilayer switching.

Cisco's gigabit netw
orking initiative spans multiple Cisco groups, including Enterprise
(which includes the former Granite Systems team), Internet Service Provider (ISP), and
Small/Medium (SMB) lines of business. Each group addresses these challenges by

leveraging Cisco te
chnologies for delivery of multiple Gigabit Ethernet products with
unparalleled performance for scaling campus networks to

gigabit rates.

The Cisco Catalyst

LAN switching architecture provides the throughput and feature set
required to scale intranet perf
ormance while it enables a smooth and stable migration of the
network core, backbone, and server connections. Cisco has developed specific mechanisms

scale link layer, Layer 2, and network
layer performance. For gigabit networking, Cisco is
evolving the

Catalyst LAN switch products to increase system throughput up to 100

while providing full routing functionality with tens


millions of packets per second
throughput, meeting the

requirement to scale campus networks from Fiber Distributed Data
face (FDDI), Fast Ethernet, or ATM


Gigabit Ethernet via a smooth migration path.


Cisco is addressing the challenges of gigabit networking through the application of leading
technology and product implementation:

Multigigabit system bandwidth
Cisco's application

specific integrated circuit
based Catalyst LAN switch architecture, scalable to more than

100 Gbps

switching capacity, delivers Layer 2 switching performance with the option of line
rate Layer 3 Forwarding
Layer 3 forwar
ding at Layer 2 performance. This feature
enables users to select the type of throughput performance they need today and
migrate performance and

functionality as needed in the future.

layer (Layer 3) forwarding and routing at gigabit line rates

NetFlow LAN Switching, which incorporates route processing, ASIC
forwarding, and

Cisco IOS software technologies, delivers line rate

performance for Layer 3 forwarding and routing of strategic protocols such as
TCP/IP while not

compromising support for multiprotocol packet forwarding.
NetFlow LAN

Switching is based on standards and seamlessly integrates into current networks,
providing network
proven scalable routing performance with industry

Application of
network services at gigabit rates
Cisco's proven route processing and
ASIC technology deliver the required network services at gigabit rates. Network
services such as security, QoS, multimedia support, mobility, and

implementation are enabled at g
igabit speeds through deployment of NetFlow LAN
Switching technology supported in Cisco's Catalyst switch products.

Monitoring and management of gigabit systems
Cisco's innovative RMON
implementation for the Catalyst LAN switch family is being extended
to deliver
monitoring, troubleshooting, and management for multigigabit systems. Unique
application of RMON I and II will deliver tools for simple, effective resolution of
gigabit networking problems.

Smooth, scalable migration
Cisco's Catalyst LAN swit
ch evolution assures a
smooth migration to gigabit networking capability. This evolution will enable
customers to migrate from FDDI, Fast Ethernet, and ATM to Gigabit Ethernet.
Technologies such as Fast EtherChannel, Gigabit Ethernet, Gigabit EtherChannel
NetFlow LAN Switching will

deliver performance improvements in manageable
increments and provide the foundation for multigigabit networking performance.
Cisco's current and future industry
leading Layer 2 and Layer 3 switching solutions
ensure robust c
oexistence and interoperability with ATM

solutions that span campus,
metropolitan, and wide


In summary, technologies including the Catalyst

LAN switch architecture, NetFlow LAN
Switching, Fast

EtherChannel technology, Gigabit Ethernet,
Gigabit EtherChannel
technology, route processing, and Cisco IOS software will be combined to deliver gigabit
rate forwarding (Layer 2 and Layer 3), routing, and application of network services at gigabit
rates to meet emerging application requirements. Th
based Catalyst switching
architecture will be scaled from current speeds to 100 Gbps. Fast EtherChannel technology,
Gigabit Ethernet, and Gigabit

EtherChannel technology deliver scalable link speeds

switching systems. NetFlow LAN Switching a
ccelerates Layer 3 forwarding to line rate
while it preserves the ability to

apply network services. Cisco IOS software provides
leading multiprotocol routing and the rich feature set found in Cisco's proven
routing systems. This unique combinatio
n of technologies ensures seamless migration from
existing network designs to gigabit networks. Finally, this set

of technologies will be
implemented in the family of Catalyst switches.

6. Technology Elements Required


Gigabit Networking

Three primary
technologies must be scaled to gigabit capabilities for mass deployment of
gigabit networking:

Layer 3 functions
Layer 3 forwarding, route processing, and application of network

Layer 2 switching
Capacity or throughput available within the L
ayer 2 switching

Link bandwidth
Bandwidth available on the physical links between switches,
routers, and servers

These technology elements must be scaled to enable product implementation that provides a
smooth migration path, leverages installed

equipment to extend useful life, and provides a
clear path toward multigigabit network performance. The Layer 3 functions can be scaled in
multiple ways that provide implementation options: one option is



system basis; the

second is by distributing

Layer 3 functionality around the network. Link bandwidth and

2 switching is scaled on a system
system basis.

Cisco will scale each of these technologies and provide a

smooth migration path via the
Catalyst LAN switching family, Cisco 7500 router

series, and the Cisco 12000 router series.
NetFlow LAN Switching is Cisco's solution to scale Layer 3 forwarding and service
application through flow caching. Key elements of NetFlow LAN Switching include Layer 3
forwarding (NetFlow forwarding), applicati
ons of

network services (NetFlow services), and
management (NetFlow management). The

following sections summarize how Cisco will
scale and

deliver the technology elements required for gigabit networks.

6.1 NetFlow LAN Switching Delivers Scalable Layer 3 F


Due to the rising levels of anywhere
everywhere communication, Layer 3 switching that
scales to tens of

millions of packets per second has become an imperative, required to speed
up both peer
peer and client/server perf
ormance on campus networks. In addition to wire
speed performance, Layer 3 switching, must provide anywhere

everywhere connections
without compromising latency, and

must meet these additional requirements for scaling


Transparently drop in
and work in the LAN switching infrastructure

Enable deployment in the LAN switched infrastructure with simple "plug

Preserve existing subnet structures

Preserve the resilience, security, and scalability of traditional routed networks

With Gigabit Ethernet on the horizon, scale to Gigabit


Cisco responds to these challenges with NetFlow LAN

Switching, which operates by
switching flows (IP conversations) at Layer 3. The first packet of a flow takes the

forwarding path. In
formation from the first packet is used to build an entry for "flow
forwarding." Subsequent packets are then Layer 3 switched at Layer 2 performance levels. In

addition, NetFlow LAN

Switching capitalizes on the flow nature of traffic to provide security
and detailed statistic measurements. By distributing NetFlow LAN

Switching technology
onto the switches, Cisco increases the aggregate Layer 3 switching capacity of the campus
intranet by an order of magnitude over current levels.

NetFlow Switching operate
s by creating a cache entry within a router, which contains the
information needed to

switch and perform appropriate services for each active flow. A flow
is defined as a sequence of packets sent from a

particular source to a particular destination.
packets are related in terms of their routing and any local handling policy they may
require. After the NetFlow Switching cache is created, packets that are identified as
belonging to an

identified flow can be switched, based on the cached information, and

appropriate services can be applied.


Extending the concept of NetFlow Switching beyond Cisco's routers, NetFlow LAN
Switching now provides network
layer switching in Cisco's Catalyst series multilayer LAN
switches at previously unmatched forwarding rates
. This switching meets the bandwidth
demands of tomorrow's next
generation backbone technologies. With development of
NetFlow LAN Switching elements such as the Route Switch Module (RSM) and NetFlow

Feature Cards, Catalyst

5000 series switches deliver m
illions of packet per second
throughput performance. NetFlow LAN Switching can be deployed at any location in the
network as

an extension to existing routing infrastructures
from the

campus backbone to
the wiring closet. With NetFlow LAN Switching, netwo
rk users can extend their use of

IOS network services without paying the performance penalty usually associated with
such processing
intensive functions. This increase in performance allows Cisco IOS network
services to be utilized from end to end wi
thin the

network and on a larger scale.

With NetFlow LAN Switching, the NetFlow feature cards silicon implements the line speed
forwarding based on

cached Layer 3 information. The RSM and Cisco IOS software continue
to perform route processing functions o

the first packet in each flow, while the

NetFlow feature cards forwards the subsequent packets in a flow) at wire speed. This wire
speed forwarding architecture allows subsequent packets to bypass the route processor,
which free route processor
bandwidth, and preserve the

added features and operating
standards of




NetFlow LAN Switching over LAN backbones uses multilayer switching silicon in the
Catalyst series of LAN switches to automatically detect flows as they are swi
tched within the
Catalyst system, and it establishes a cut
through path whenever a flow is detected. Cisco IOS
features of Catalyst LAN switches enable the switch to discover router information that is
required for detecting candidate "flows," which, in tu
rn, enable a RSM to inform the switch
of routing or policy changes. Cisco IOS router software signals the NetFlow Feature Card to
purge stale forwarding entries, for

example, when an access list changes, and it enables
switches to respond to network failur
es or topology changes

to provide fast convergence



Because the implementation of NetFlow LAN Switching in Catalyst switches is based on
Cisco ASIC technology, the Layer 3 throughput can be scaled from a few millions of packets
per second to tens of mil
lions of packets per second. When combined with the RSM
processing power, based on

end Cisco RSP technology, and Cisco's next

gigabit route processing engines, a complete solution for multigigabit Layer 3 forwarding
and application of netw
ork services is delivered.

Cisco's solution to scaling Layer 3 forwarding and services employs multiple approaches.

These approaches provide network architects a variety of options to achieve a

migration to multigigabit performance. The key elemen
ts of Cisco's NetFlow LAN Switching
architecture for

scaling are: route processing, multilayer switching, and

NetFlow silicon
(ASICs) implemented on the NetFlow Feature Cards. Each of these elements can be scaled
within a

system or around the network..Firs
t, the Catalyst RSM delivers route processing and
Layer 3 forwarding for deployment in Catalyst multilayer LAN switches. This feature
enables the RSM to be placed at locations in the network that require increased Layer 3
functionality for scaling performa
nce. When more throughput is required, increase the
number of RSMs in the Catalyst switch.

Next, NetFlow ASICs are deployed via the NetFlow Feature Cards with the RSM and
provide accelerated Layer 3 forwarding, millions of packets per second, at campus
locations that require high
performance Layer 3 switching. As route processing performance
and ASIC technology increases these improvements will be incorporated into Catalyst LAN
switches to deliver tens


millions of packets
second performanc


The first steps towards gigabit networking have been taken by exploring the capabilities of
various media which can support gigabits of bandwidth. There is consider
able amount of
research being done in various gigabit testbeds, but still gigabit as field has a long way to go
before becoming mature. One of the main issues is what protocol should gigabit networks
run on. It is interesting to note here that, traditional

protocols like IP, TCP with various
modifications and extensions have been shown to support gigabit bandwidth. But should we
stick with these are not is a question yet to be answered. Most of these protocols have not
been verified because they have a huge

number of states. Protocol verification is an
important issue which has to answered by gigabit networking since at such high speed
something can go wrong very fast. Some of still unexplored issues in gigabit networking is
Network management. It is unclear

whether the existing network management will still work
at gigabit speeds. Encryptions of data takes some time, so fast encryptions algorithms (like
DES, Digitial Encryption Standard) should be developed if authentication is to be supported
by gigabit net

In most of the operating systems since the network interface is not well integrated, there is lot
of redundant copying of data. So operating systems face the challenge of changing to support
better network interfaces. This will become critical in s
upporting real
time systems. Some
work has already been done in this area with improvements in processor speeds, better
scheduling and caching techniques.

Traffic modeling is another area where work has to done. It has been recently shown that
Ethernet tr
affic is Self
Similar (fractal) in nature. It has to verified whether this hold for
gigabit networks or come up with new models for traffic analysis.

Will the network world stop progressing after gigabit networks or will terabits network be
developed ? It

is already been known that the potential of a single fiber is in the order of
terabits. We can envisage terabit speed networks were a parallel array of high speed
computers are connected. But will there be terabit applications to run on it? As it happens

most of the time an application will be found whenever the technology has been invented. So
we can expect to be working with terabit networks in the future.


1. Gigabit Networkin
g (Craig Partridge ).

2. White paper on Gigabit Networking by Cisco Systems.

3. Research paper by Prof.Raj Jain,Ohio State university on Gigabit Networking.