Gigabit Ethernet in Context withOtherLAN Technologies
Gigabit Ethernet Networking
Author: David Cunningham; Bill Lane
As stated in Chapter 13, “Upgrading Ethernet
LANs: System and Topology Considerations,” Gigabit Ethernet
can be used to upgrade both Ethernet and non-Ethernet local area networks
(LANs). Chapter 13 discussed upgrading Ethernet
LANs. This chapter briefly reviews some non-Ethernet technologies currently
deployed in LANs and then considers the use of Gigabit Ethernet as an upgrade.
There are several reasons why you also should consider upgrading non-Ethernet
networks with Gigabit Ethernet technologies:
Different divisions within the same organization may have
initially chosen to implement different LAN technologies and protocols.
Gigabit Ethernet can be the lowest cost long-term alternative
for upgrading, even within a non-Ethernet environment. Over 80 percent of
the nodes in the world's LANs use Ethernet technology. Furthermore, because
this is a highly competitive market, costs can be expected to continue their
The Ethernet protocol is typically less complex, more easily
understood, and simpler to implement than competing protocols.
Ethernet can now provide previously unavailable Class of Service
It is widely recognized that the first major application to be addressed
by LAN equipment upgrades is the need for increased bandwidth between workgroup,
backbone, and campus networking equipment. But, the introduction of new data
types requiring CoS and QoS features means that raw bandwidth alone is not
sufficient. For this reason, all new switch-based products, whether packet-
or circuit-based, already have or are introducing techniques that can provide
CoS and QoS features, at least to some degree.
This chapter begins with a review of commonly installed LAN technologies,
then continues with a discussion of the higher layer protocols required to
both support IP-based traffic on connection-oriented networking technology
and to support QoS or CoS over IP-based networks. The chapter concludes with
a discussion of the suitability of Gigabit Ethernet as an upgrade for each
of the LAN technologies discussed.
As you should be aware by now, LAN equipment
is installed in a hierarchical manner. The hierarchy is related to the building
wiring standard levels for horizontal cabling (for connection to desktops),
building backbone cabling (for interconnection of workgroup LAN equipment),
and campus backbones (for interconnection of building LAN equipment). This
hierarchy has evolved because it eases the burdens associated with network
design, network management, and fault isolation of both LAN equipment and
the supporting cabling infrastructure.
Table 14.1 summarizes some of the
important features of commonly installed Layer 2 LAN technologies. ATM is
the only connection-oriented LAN technology listed in Table
14.1. All the others are connectionless (packet-based) technologies.
Table 14.1. Multimode Fiber (MMF) Link Lengths for Commonly installed Layer 2 LANs
|Ethernet||100||Shared||Desktop and Wiring closets||412|
|Ethernet||100||Switched||Backbone and Desktop||2,000|
|Token Ring||16||Shared and Switched||Desktop||2,000|
|Token Ring||100||Switched||Backbone and Desktop||2,000|
|FDDI||100||Shared and Switched||Backbone||2,000|
|100VG||100||Shared and Switched||Desktop and Backbone||2,000|
a This distance
could be increased to 1,600 m if mode-conditioned 1000BASE-LX Gigabit Ethernet
optical transceivers were used for this application.
14.1 states the maximum multimode fiber link length, you should remember
that practically all desktop links use Category-5 UTP cable. Also, as noted
in Chapter 13, if single-mode optical fiber is
used for switched links, the maximum link length could be very long indeedup
to 110 kmwith commercially available, specialized, non-standards-based,
single-mode link extenders.
Token Ring is a deterministic protocol
that was developed in the early 1980s to avoid the random packet collisions
that occur in Ethernet. Token Ring has been standardized as IEEE 802.5. Originally
the network topology was a physical ring. However, today many Token Ring networks
have their wiring installed in the normal building wiring tree or star topology.
In the ring configuration, both tokens (permission to send) and information
packets are transmitted between successive stations or nodes on the ring.
Only one token is allowed on the ring and only the node holding the token
is allowed to transmit a data packet. After the sender has finished transmitting
a data packet, it generates a new token that is then circulated around the
ring. The token will be captured by the next node on the ring needing to send
The Token Ring frame
format is different from the Ethernet format both in how the information is
encoded and in the field definitions for the frame format. The data field
length can be up to 4,500 bytes, and two additional fields provide control
information that is not available in a basic Ethernet frame:
A 1-byte access control field in the frame header contains 3 bits
for packet priority designation, 3 bits to indicate a token reservation and
its priority level, plus a token bit to differentiate a data/command frame
from a token.
A 1-byte end delimiter identifies the end of the frame and contains control
bits to indicate whether the frame was damaged in transmission and whether
this frame is the last frame in a logical sequence.
Currently available Token Ring products operate at 4, 16, and 100 Mbps
(100 Mbps Token Ring is in the final stages of standardization; also, a task
force has been set up to draft a Gigabit Token Ring standard). Both shared
and switched versions of Token Ring are available.
The Fiber Distributed Data Interface (FDDI) is a Token Ring protocol
similar to IEEE 802.5 that operates at 100 Mbps. FDDI was developed in the
1980s as a high speed backbone and desktop replacement for IEEE 802.5 and
was standardized through ANSI by the X3T9.5 working group. Originally, FDDI
was designed as a fiber optic network; however, copper UTP and STP versions
were developed. The FDDI physical layers were reused by the IEEE 802.3 committee
for 100BASE-FX and TX Fast Ethernet. The packet format used by FDDI is similar
to that of Token Ring.
FDDI actually uses two counter-rotating rings to provide a high degree
of fault tolerance. Comprehensive station management was developed as part
of the standard and FDDI's main role has been as a highly reliable backbone
LAN. Both shared and switched FDDI products are available. A higher data rate
version of FDDI has not been developed.
uses the Demand Priority Access Method (DPAM), a deterministic
protocol that was developed in the early 1990s as a high speed upgrade for
both 10BASE-T and 4 Mbps and 16 Mbps Token Ring LANs. As such, a single DPAM-based
network can operate in either an Ethernet mode, where it transmits Ethernet
frames, or a Token Ring mode, where it transmits Token Ring frames, but not
in both modes at the same time. DPAM also provides two levels of packet priority.
DPAM was standardized as IEEE 802.12.
100VG-AnyLAN was specifically designed to operate over the lowest grade
UTP cable used by 10BASE-T: 25 pair, voice grade UTP cabling. Because of this,
it has a significant operational margin on better grades of UTP, STP, and
optical fiber cabling. With shared media, 100VG-AnyLAN obeys the same topology
rules as 10BASE-T. Both shared and switched versions of 100VG-AnyLAN are available.
There was an initial effort to develop a Gigabit version of IEEE 802.12;
however, with the advent of switch-based networks, it was recognized that
Gigabit Ethernet would be a very effective upgrade for 100VG-AnyLAN. Gigabit
activity within IEEE 802.12 was terminated, and many of the contributors from
that committee transferred to the IEEE 802.3z committee, where they played
key roles in the development of Gigabit Ethernet.
Transfer Mode (ATM) is probably the best known alternative to Gigabit Ethernet.
For this reason, this section provides more coverage than is provided for
the other LAN examples. ATM was developed by the telecommunications industry
as part of the Broadband Integrated Services Digital Network
(BISDN). It provides a means to integrate various data types having disparate
transmission requirements, such as computer data traffic, voice traffic, and
video traffic. In contrast to the examples considered thus far (Ethernet,
Token Ring, FDDI, and 100VG-AnyLAN), ATM is connection-oriented.
ATM transmits a continuous stream of small fixed length packets called cells.
An ATM cell is only 53 bytes long, consisting of a 5-byte header and a 48-byte
payload field. ATM cells were purposely kept short to reduce transmission
delay variations and to enable efficient switching.
The header contains the control and next destination routing information
for the cell, including:
A generic flow control (GFC) field that is seldom used, but typically
set to a default value
A virtual path identifier (VPI) that is used with the VCI to identify
the next route destination of the cell
A virtual channel identifier (VCI) that is used with the VPI
type (PT) field that indicates whether the cell contains data or control information
A congestion loss priority (CLP) bit that indicates how the cell
should be treated with respect to other cells in cases of network congestion
A header error control (HEC) field containing a checksum for the
of the ATM layer is to provide transparent transfer of cells across pre-established
network connections according to a pre-arranged traffic contract. ATM systems
generally use a Synchronous Optical Network (SONET) physical layer. When this
is the case, the ATM layer must insert unassigned (idle) cells into the stream
of cells during times when there is no data to transmit. The result is a continuous
stream of cells, a process called cell rate decoupling.
The VPI and VCI are used by ATM switches, which form part of the ATM
layer, to manage connections across the ATM network. Because ATM is a connection-oriented
protocol, end-to-end signaling must be used to establish a virtual circuit
before any data is transmitted. The signaling establishes the path through
the ATM switching network and any required QoS agreements.
Another important part of the ATM layer is the ATM
adaptation layer (AAL). The ATM adaptation layer usually resides in end stations.
The AAL layer in the transmitting station segments frames received from higher
protocol layers into cells for transmission; the peer AAL layer in the receiving
station collects the cells and reassembles them into frames for forwarding
to the higher protocol layers. Various types of AAL have been standardized
for use with different traffic types. AAL5 is commonly used for packet-based
in sections 14.1.1-14.1.4 was limited to issues relating to layer 1
(the physical layer) and layer 2 (the data link layer). For a more complete
comparison of Gigabit Ethernet with the other networking options, you need
to consider some of the higher layers. This discussion begins with a brief
review of the most commonly used LAN model: the TCP/IP suite. The TCP/IP protocol
suite doesn't have the same layered model as the OSI reference model. Roughly
speaking, however, IP can be associated with the network layer and TCP with
the transport layer, although TCP also includes some of the OSI session layer
functionality. In Ethernet implementations, the bottom two layers consist
of the IEEE 802.2 LLC and IEEE 802.3 MAC (data link layer) and the IEEE 802.3
physical layer, as shown in Table 14.2.
Because IP is a connectionless protocol, no end-to-end signaling
or control information needs to be exchanged before packet transmission begins.
Also, because IP only provides a best effort service,
no QoS or CoS is provided. TCP is the transport layer protocol of the TCP/IP
suite and it provides connection-oriented reliable transmission (see Chapter 4, “Gigabit Repeaters, Bridges, Routers, and
Switches”). TCP is a protocol that is best suited to data communications
applications. TCP/IP is not a good choice for real-time video transmission.
An IP-based network has two types of nodes: hosts
and routers. A router is a node with at least two network
interface connections and is capable of forwarding packets from one network
to another network. A host is simply a node that is not a router.
Three other important protocols that support the IP protocol are
the Internet Control Message Protocol (ICMP), the Internet Group Management
Protocol (IGMP), and the Simple Network Management Protocol (SNMP), all of
which are standardized. ICMP is used mainly for fault diagnosis, error reporting,
and control messages. IGMP is used to form and manage multicast groups. SNMP
enables network administrators to manage the network.
In an IP-based internet, routes are calculated by the network routers
in a distributed fashion. The IP protocol does not define how the routers
should do this. However, there are standard routing
protocols for this purpose.Some well-known examples of standards-based
IP routing protocols include the following:
Routing Information Protocol (RIP)
Open Shortest Path First (OSPF)
Border Gateway Protocol (BGP)
QoS and CoS in IP-Based Networks
Until the advent of multilayer switching,
routing was performed by general purpose processors under software control.
But because software routines are typically slower than ASIC hardware routines,
software-based routers can be congestion points in a network. As such, software-based
routers could make it difficult to provide QoS/CoS service; however, IP doesn't
provide either QoS or CoS anyway.
As section 13.2 in Chapter 13 stated, the
Internet Engineering Task Force (IETF) has been working for some time to define
standards-based mechanisms and protocols for providing QoS and CoSespecially
for switch-based networks. They are also finalizing the Resource Reservation
Protocol (RSVP), which will enable networks to reserve dedicated bandwidth.
In addition, the IP-based multicast backbone (Mbone),
which was based on the Real-time Transport Protocol (RTP), has proven that
real-time, multicast video conferencing is possible over the Internetthe
Mbone was originally designed to broadcast the meetings of the IETF.
Even before these standards are finalized, many manufacturers of high-end
multilayer switches have implemented the common IP-based routing algorithms
(RIP, OSPF, BGP, and so forth) in hardware to produce routing-switches.
These routing switches can process packets at the line rate of each port (even
for Gigabit Ethernet!). Usually, each port has dedicated routing hardware
so that packets entering the port can either be switched, based on layer 2
information or, if need be, routed, based on layer 3 information. Many of
these routing switches can also provide CoS and QoS features based on some
or all of the following: IEEE 802.1p/Q, IP datagram Type of Service (ToS),
and the source or destination TCP/UDP port number. But not all protocols are
routed in hardware. Those that aren't are usually forwarded to a software-based
router. However, the network performance for these other protocols is also
greatly improved because the software-based routers are only required to deal
with non-IP traffic.
Several manufacturers are using cut-through
switches to gain performance improvement, compared to software routing. Cut-through
switches usually gain performance improvements by switching based on the examination
of only the packet header of each packet in a transmission. However, some
routers gain a further increase in performance by only examining the first
few packets of an identified transmission. Several of the schemes used for
IP cut-through routing actually involve using a switch with an underlying
ATM-based switching fabric. Cut-through switch-routers tend to use proprietary
methods, although standards are beginning to emerge.
As stated in section 14.1.4, ATM is a connection-oriented switch-based
layer 2 technology that requires end-to-end signaling before any data packets
are transmitted. However, the end-to-end signaling can also be used to set
up connections with guaranteed levels of QoS and CoS.
While ATM is ideal for voice and video applications, it has some difficulties
when transmitting IP-based data. These difficulties arise because IP was predicated
on a connectionless service model. Broadcast and multicast traffic is commonly
used in Ethernet and IP-based networks for functions such as address discovery
and routing table updates. ATM requires additional higher layer protocol support
to cope with these broadcast and multicast requirements, but because over
80 percent of the world's network connections are through Ethernet ports,
support of IP traffic is no longer optional.
Even before the growth of the World Wide Web, it would have been difficult
to change the software associated with networking to a connection-oriented
model simply because most applications assume a connectionless network model.
Table 14.3 compares the TCP/IP/ATM
layer-2 protocol stack to the OSI model.
Table 14.3. The Relationship of the TCP/IP/ATM Protocol Stack to the OSI Model
a MPOA is Multiprotocol
over ATM, LANE is LAN Emulation, MPLS is Multiprotocol label switching, and
RFC 1477 is the IETF's classical IP-over-ATM standard. To simplify the table,
ATM end-to-end signaling protocols are not shown.
|OSI Layer||OSI Layer Name||Protocol Implementation|
|3||Network||IP, LANE, MPOA, MPLS, or RFC
Link||ATM AAL5 and ATM switching layer|
So, it seems we've discovered a Catch-22 situation:
ATM has native QoS, but requires complex higher layer protocol
support for traditional IP-based traffic.
Ethernet is ideal for IP-based traffic, but requires complex
higher layer protocol support in order to provide CoS for emergent multimedia
In general, there is no solution to this Catch-22 situation. The best
that can be done is to use the appropriate technology for the situation at
hand. However, it is now widely recognized that Ethernet technology is best
suited for the LAN, and ATM is best suited for the WAN; it is expected that
both will compete for use in MANs. The next few sections introduce some of
the higher layer protocols commonly used to support IP traffic on ATM-based
The LAN emulation (LANE)
protocol defines the operation of a single emulated LAN
(ELAN). LANE uses ATM connection-oriented switches in a way that emulates
a traditional connectionless LAN. For example, in the case of an Ethernet
emulated LAN, the LANE interface behaves the same as a standard Ethernet layer
2/3 interface. This means that all data sent through the ATM network consists
of ATM-encapsulated Ethernet packets.
Multiple ELANs can be simultaneously operated on a common ATM network.
However, a separate ELAN must be dedicated to each traditional LAN protocol
in operation. LANE was intended to accelerate the introduction of ATM networks.
A disadvantage of LANE is that because it emulates a traditional connectionless
layer 2/3 LAN interface it cannot use ATM's native QoS features.
LANE is based on a client-server model. The LAN emulation
client (LEC) is an end-station function that performs data forwarding, address
resolution, and registering of the MAC address with the LAN emulation server.
For traditional LANs, the LEC also provides a standard layer 2/3 network interface
to higher layer protocols. If an ATM end system is connected to multiple ELANs,
there must be a separate LEC for each ELAN supported.
LANE requires three servers as follows:
LAN Emulation Server (LES). One LES is required
for each ELAN. It is a central point that controls the admission of LECS to
the ELAN. It is also responsible for registering client MAC and ATM addresses,
and for address resolution.
LAN Emulation Configuration Server (LECS). One
LECS serves all ELANs within an administrative domain. It maintains a database
of the LECs and the ELANs to which they are registered. When queried by an
LEC, the LECs will respond with the ATM address of the LES that serves the
ELAN appropriate to that LEC.
Broadcast and Unknown Server (BUS). Each LEC
is associated with one BUS per ELAN. The BUS is responsible for delivering
broadcast frames and multicast frames. Additionally, unicast frames containing
addresses that are unregistered or as yet unresolved are delivered by the
illustrated in Figure
14.1, the LANE protocol is deployed in two types of ATM equipment:
ATM network interface cards (NICs) and internetworking equipmenta
layer2 LAN switch is shown in the figure. Because an ELAN emulates a specific
traditional LAN (Ethernet, Token Ring, FDDI, and so on), it can support
various higher layer protocols such as IP, IPX, SNA/APPN, and NetBIOS.
14.1. Example of How LANE Protocol Can Be Implemented in ATM Network
Because IP over ATM as specified in RFC 1477
does not change the basic nature of the IPprotocol, it is called Classical
IP over ATM. Classical IP over ATM has the following characteristics:
It is based on the concept of a logical IP subnetwork (LIS).
An LIS contains hosts and routers having the same IP subnetwork
mask (the bits of an IP address that are being used for the common subnetwork
address) and the same subnetwork address.
Any two hosts of the same LIS may communicate directly using
virtual channels (VCs).
Hosts from different LISs may only communicate through a router.
The maximum transmission unit (MTU) is the same for VCs in
an LIS and is a maximum of 9,180 bytes.
Within a single LIS, IP addresses are resolved to ATM addresses
by an ATM address resolution protocol (ATMARP)
server. The inverse resolution is also performed by the inverse ATMARP on
Classical IP over ATM has the obvious advantage that it allows IP data
to be routed over an ATM LAN or WAN. However, its main disadvantages are that
it only supports IP routing, it has no multicast support, ATM's intrinsic
QoS properties may be lost passing through routers, and it is difficult to
scale to a large number of LISs.
In an MPOA environment, connectionless LAN traffic is directed to shortcut
ATM paths. For this reason, MPOA is classed as a shortcut
or cut-through routing scheme. MPOA uses three technologies
LANE is basically
used for configuration purposes, NHRP is used to determine shortcut paths
through the ATM switch-based network, and the virtual router concept allows
partitioning of routing functions among various physically separated elements
of the network.
The routing calculations are performed by the MPOA Servers (MPSs), while
traffic forwarding is the responsibility of the MPOA Clients (MPCs) and the
ATM switching layer. An MPS could be a standalone ATM attached route server,
or it may reside within an ATM switch-router. MPCs reside in network edge
devices and in ATM attached hosts. Traditional (non-ATM) hosts connect to
the ATM network through the edge devices that support traditional LANs on
the LAN side of the device and ATM on the backbone side.
Native ATM devices may be directly connected to the ATM layer. The MPOA
routers or route servers within the ATM network provide routing information
to the edge devices. The edge devices transmit data from their connected hosts
to other edge devices using the shortcut ATM layer paths identified by the
route servers. Again, although QoS is available to all ATM-attached devices,
traditional connectionless-oriented LAN devices may not be able to take full
advantage of this feature.
MPLS is a form
of tag switching. Tag switching was introduced in Chapter
4 and will not be discussed here. In an ATM context, VCs would be identified
with particular labels ortags.
One of the primary architects of OpenCable, Michael
Adams, explains the key concepts of this initiative in his book
Broadband, Second Edition
by George Abe
Introduces the topics surrounding high-speed networks
to the home. It is written for anyone seeking a broad-based familiarity
with the issues of residential broadband (RBB) including product
developers, engineers, network designers, business people, professionals
in legal and regulatory positions, and industry analysts.