ETHERNET SWITCH

A multiport device based on bridging technologies that is mainly used
to segment Ethernet networks.

Overview

Often simply called switches (when referring implicitly to Ethernet
networking), these devices are used to enhance the performance of
Ethernet networks. An Ethernet switch basically resembles a hub, and
consists of a box with a number of RJ-45 jacks on the front to provide
ports for network connections. Inside, however, are advanced
electronics that generally make switches more costly than hubs.

In a hub, a packet entering one port is regenerated and forwarded to
every other port. While the packet is being forwarded, no other port
can receive packets, so a hub can be thought of as a shared-media
device in which all ports are connected using a shared bus. If a
collision occurs on a hub-based Ethernet network, no port on the hub
can receive traffic until the collision is resolved.  The set of
stations connected to a hub is thus called a collision domain. On the
average, if a 10-megabits-persecond (Mbps) Ethernet hub has 10 ports,
each port effectively gets one-tenth of the total bandwidth, or 1
Mbps. In reality this can be much worse, however, for a single station
actively transferring files can consume a large percentage of the
available 10 Mbps bandwidth of the hub, leaving other stations starved
for bandwidth for their communications. Hub-based Ethernet networks
are thus based on contention in which every station must fight for its
share of bandwidth.

If you connect several hubs and their networks by uplinking them to a
main hub, the situation only gets worse because the new larger network
remains a single collision domain. With the increased number of nodes
on the network, however, more collisions are likely to occur and
network traffic congestion can result, slowing the network to a crawl.

The solution to this congestion problem is to strategically use
Ethernet switches in place of, or in addition to, hubs. When a packet
enters a port of an Ethernet switch, the switch looks at the frame.s
destination address, compares it to a table of address-to-port
mappings maintained internally by the switch, internally establishes a
temporary internal logical connection between the incoming port where
the packet arrived and the outgoing port where the packet is destined,
and forwards the packet along this internal connection to its
destination.  Only the port where the packet arrived and the
destination port are involved in this process; all other ports on the
switch have no part in the connection.  The result of this process is
that each port on the switch corresponds to an individual collision
domain, and network congestion is therefore avoided. Thus, if a
10-Mbps Ethernet switch has 10 ports, each port effectively gets the
entire bandwidth of 10 Mbps, and to an incoming packet the switch.s
port appears to provide a dedicated connection to the destination node
on the network.  In other words, replacing switches with hubs does not
add bandwidth to your network. Instead, it reduces the size of
collision domains to allow bandwidth to be used more efficiently. If
12 stations are connected to a 12- port 10 Mbps hub, the maximum
bandwidth any one station can theoretically use is 10 Mbps, but if a
station were actually using this much bandwidth, all remaining
stations connected to the hub (except for the station being
communicated with) would have zero bandwidth available to them. If a
12-port, 10 Mbps switch were used instead of the hub, however, all 12
stations could participate in 10 Mbps conversations simultaneously,
which means the switching fabric supports 12 ports = 6 communications
x 2 ports/communication = 6 x 10 Mbps = 60 Mbps throughput. However,
the total bandwidth available to any one station is still only 10
Mbps, no more than for the hub. The difference is that this 10 Mbps
bandwidth is guaranteed to be always available for each station!

Architecture

A basic Ethernet switch operates at Layer 2 (the data link layer) of
the Open Systems Interconnection (OSI) model. Like a bridge, an
Ethernet switch is a smart device that can learn the media access
control (MAC) address of each connected station by listening to
network traffic. The switch builds an internal table listing the MAC
address of each port and consults this table when it needs to forward
incoming packets.

When an Ethernet frame arrives at a port, the destination MAC address
is read from the first 64 bits of the frame. This destination address
is then found in the switch.s internal address table to determine the
correct destination port for the frame. Once this is determined, the
switching fabric (a mesh-like connection) inside the switch
establishes a temporary logical connection between the incoming and
destination ports, forwards the frame, and then tears down the
connection. Ethernet switches are also capable of establishing
multiple internal logical connections simultaneously between differ-
ent pairs of ports. The result is that each port receives the switch.s
full dedicated bandwidth at all times, giving Ethernet switches
intrinsically much more bandwidth than shared hubs.

The actual mechanism by which switching (the forwarding of the packet
between the ports) occurs divides Ethernet switches into two general
device classes:

. Store-and-forward switches: This type waits until the entire
incoming frame arrives, buffers the frame, reads the destination
address from the buffered frame, performs cyclical redundancy check
(CRC) error checking to verify that the frame is valid, and either
forwards the frame to the destination port using the switching fabric
or drops the frame if it represents a runt, jabber, or some other
invalid frame. Store-and-forward is the same method by which most
bridges work. The advantages of this method are that bad frames are
eliminated and collisions are handled well, while the disadvantage is
that it suffers from high latency (delay) because the frame must be
entirely read and buffered before being processed. Nevertheless, most
switches used today are of the store-and-forward type, and
Transmission Control Protocol/ Internet Protocol (TCP/IP) is generally
well able to handle any additional latency that arises in networks due
to use of these switches.

. Cut-through switches: This type reads the frame as it comes into the
receiving port, and after the source and destination addresses are
read from the first 64 bits, the switch then checks the internal
address table and immediately forwards the frame to the correct
destination port. Error checking is not performed by cut-through
switches, since to perform this check the CRC byte would have to be
read first, and this is located at the end of the frame. As a result
of not performing error checking, cut-through switches reduce latency
and accelerate performance over store-and-forward switches, but the
result is that collisions can affect overall communications and bad
frames are passed rather than dropped. The earliest Ethernet switches
were of the cut-through type and found particular application in
NetWare 3.x networks where Internetwork Packet Exchange/ Sequenced
Packet Exchange (IPX/SPX) was less tolerant to delay than Internet
Protocol (IP). Limitations in bridging electronics also made
cut-through switching appealing in the earliest implementations, but
advances in electronics have made storeand- forward switching more
appealing. The first commercial Ethernet switches were cut-through
switches developed in 1990 by Kalpana, which is now a subsidiary of
Cisco Systems.

Implementation

Ethernet switching (or switched Ethernet) can be implemented in
various ways depending on the OSI layers at which the switches
operate. These include

. Layer-2 switch: This basic Ethernet switch operates at the OSI
data-link layer (Layer 2) and is based on bridging technologies as
explained above. Layer-2 switches establish logical connections
between ports based on MAC addresses and are generally used for
segmenting an existing network into smaller collision domains to
improve performance.

. Layer-3 switch: This type operates at the OSI network layer (or
Layer 3) and is based on routing technologies.  Layer-3 switches
establish logical connections between ports based on network addresses
such as IP addresses. You can use Layer-3 switches instead of routers
for connecting different networks into an internetwork, and they
generally perform better than routers due to their enhanced
electronics. Layer-3 switches are sometimes called routing switches,
multilayer switches, or sometimes simply routers.

. Layer-4 switch: This is basically an enhanced Layer-3 switch that
has the additional capability of examining the source and destination
ports of Transmission Control Protocol (TCP) and User Datagram
Protocol (UDP) connections and making forwarding decisions based on
this information.  They are called Layer-4 switches because they
examine OSI transport layer (Layer 4) information within frames and
determine how to forward the frames based on this information.

. Layer-7 switch: In this type of device the entire frame is
unpackaged to determine OSI application layer (Layer 7) information
such as what application layer protocol is being used, such as
Hypertext Transfer Protocol (HTTP) or File Transfer Protocol
(FTP). Frames can then be forwarded or dropped based on this
information.

. Multilayer switch: This is simply a switch that uses unpackaged
frame information from several OSI layers to determine how to forward
frames.  For example, most Layer-4 switches are actually Layer-3/4
switches since they operate at both layers.  Ethernet switches are
also distinguished in other ways, such as by the number of ports they
have, whether they operate in half-duplex or full-duplex mode, their
transmission speed (for example, 10 Mbps, 1/100 Mbps, or 100/1000
Mbps), ports for connectivity with high-speed Fiber Distributed Data
Interface (FDDI) backbones, and so on. Advanced features can include
Simple Network Management Protocol (SNMP), out-of-band management
(OBM), and custom packet filtering.

Uses

Ethernet switches have two basic uses: segmenting networks to improve
performance and interconnecting networks of different speeds. The most
common use is network segmentation, and introducing Ethernet switches
will provide the most obvious benefit for parts of your network where
the most contention occurs. For example, if you have several heavily
used servers on the same hub-based local area network (LAN) as
clients, the clients have to contend for use of the servers and the
result is poor performance. To improve things, segment your LAN into
several collision domains, one for each hub and one for each group of
clients, as shown in the figure. Now several clients can connect to
different servers simultaneously and receive the full throughput of
those servers. A good rule of thumb for deciding whether to use
switches to segment your existing network is that switches can improve
your network.s performance if the current network utilization level is
higher than 35 percent or if collisions are running at more than 10
percent.

The second main use for switches is to connect fast workgroup hubs to
slower hubs on a network. Again, the main hub is replaced by an
Ethernet switch, typically 10/100 or 100/1000 Mbps, and the
performance of stations on the fast hub is no longer hindered by the
presence of the slower parts of the network. Another related use for
Ethernet switches is to connect 100-Mbps Ethernet .islands. to an
existing 10-Mbps Ethernet LAN. Simply use a 10/100-Mbps Ethernet
switch with two ports to connect them. You can also connect two LANs
several kilometers apart by using two Ethernet switches, both having
one 100BaseT port and one 100BaseFX port. Connect the switches to the
LANs, and then connect a fiber-optic cable between the FX ports.

If users in a department have high bandwidth needs, such as those
running computer-aided design (CAD) or multimedia applications,
consider replacing their workgroup hub with an Ethernet switch, or if
the number of users is small, connect their stations directly to the
main Ethernet switch.

When purchasing Ethernet switches, make sure they have Remote
Monitoring (RMON) agents built into each port, which will considerably
ease remote network troubleshooting.

Marketplace

Ethernet switches are made all shapes and sizes from dozens of
different vendors. They vary from small 12-port 10/100 workgroup
switches to modular 1 Gbps backbone switches supporting Asynchronous
Transfer Mode (ATM) and Synchronous Optical Network (SONET)
connectivity. Probably the most popular 10/ 100 Ethernet switches are
those of the Cisco Catalyst 3500 Series XL, which by some counts are
deployed twice as often as any other type of similar switch from other
vendors. A popular enterprise switch used for collapsed backbones is
the Big Iron 4000 switch from Foundry Networks. Another widely
deployed backbone switch is the Hewlett-Packard 9304.

Issues

Although Ethernet switches relieve traffic congestion by segmenting
collision domains, they do have some disadvantages:

. They are generally several times more expensive than hubs of the
same speed.

. Networks involving switches are more difficult to monitor and
troubleshoot.

Ethernet switches should generally be implemented judiciously within
Ethernet networks. Simply replacing every hub with a switch is an
unnecessary expense that brings negligible performance enhancement
over just replacing a few key hubs with switches.