1494 Chapter 39
stations respond to packets received addressed to them,
when the response is sent the switch learns which port
that address is on. Thereafter packets addressed to that
station are only sent out the correct port.
If a given MAC address is found on a different port
than was contained in the LUT, the LUT is corrected. If
no packet is received from a given MAC address within
a timeout window of perhaps 5 minutes, its entry in the
LUT is deleted. These characteristics allow the switch
to adapt and learn as network changes are made.
Packets intended to go out a given port are never
allowed to collide inside the switch. Instead each
outgoing packet is stored in a first in first out (FIFO)
buffer memory assigned to a given port, and transmitted
one at a time out the port.
While most data passing through a switch behaves as
described above, there is one type of packet that does
not. Most data packets are addressed to a specific desti-
nation MAC address. This is called unicast addressing.
There is a specific address called the multicast or broad-
cast address. Packets with this address in their destina-
tion field are sent to all stations. Therefore, these
packets are sent out all ports of a switch except the port
they came in on.
Switches are not the shared media of the early
coaxial cable Ethernet varieties, or the newer repeater
hubs. Instead by storing the packets, examining the
addresses, selectively passing the packets on, and FIFO
buffering the outputs, they break the network diameter
limitation.
Switches have another difference from repeater
hubs. Repeater hubs and stations connected to them
operate in half duplex mode. In other words a given
station can only receive or transmit at different times. If
a station that is transmitting in half duplex mode sees a
received signal, that tells it a collision has occurred.
Since switches store and buffer the packets, they can
operate in full duplex mode with other switches or with
stations that can operate full duplex. When a station is
connected to a switch in full duplex mode it can receive
at the same time as it transmits and know that a collision
can’t occur since it is talking to a full duplex device that
does not allow collisions to occur internally.
Full duplex operation has the added benefit of
doubling the communications bandwidth over a half
duplex link. A half duplex fast Ethernet connection has
100 MBit/s of available bandwidth that must be split
and shared between the packets going each direction on
that link. This is because if packets were going both
directions at once, that, by definition, would be a colli-
sion. A full duplex link, on the other hand, has no
problem allowing packets to flow in both directions at
once, so a fast Ethernet link has 100 MBit/s capability
in each direction.
Of course a repeater hub-based fast Ethernet network
has only 100 MBit/s of total bandwidth available for the
entire network since it uses a shared media. A network
based entirely on fast Ethernet switches has 100 MBit/s
of bandwidth available in each direction on each link
that makes up the network, assuming that all the stations
are capable of full duplex operation. When you combine
no collisions with full duplex operation, a switched
network can run much faster than a repeater hub-based
network.
The internal packet routing function inside a switch
is called the switch fabric or switch cloud. Switches that
contain enough packet routing capability in their cloud
to never run out of internal bandwidth, even if all ports
are receiving the maximum possible amount of data, are
known as nonblocking switches.
Proper Ethernet network design includes ensuring
that no packet may go through more than seven
switches on its way from the source to the destination.
When switches were first introduced their expense
limited their application to the few situations that
required their capabilities. Today the price of switches
has come down until they are hardly any more expen-
sive than repeater hubs. As a result the repeater hub is
becoming a vanishing part of Ethernet history.39.11.7 Ethernet Connection NegotiationIt is important to understand how different Ethernet
devices negotiate connections between themselves in
order to understand why some combinations of devices
will work and others won’t.
If a 10 MBit/s Ethernet device is not transmitting
data, its output stops. After a period of no data trans-
missions, it will begin sending normal link pulses
(NLPs). These allow the device at the other end of the
link to know that the connection is still good, and it
serves to identify the device as a 10 MBit/s device.
100 MBit/s devices on the other hand always send a
signal even when no data is being transmitted. This
signal is called a carrier, and serves to identify the
device as a 100 MBit/s device.
10/100 Ethernet devices often use a technique called
autonegotiation to establish the capabilities of the
device at the other end of the link, before the link is
established. This process determines if the other device
is capable of full or half duplex operation, and if it can
connect at 10 MBit/s, 100 MBit/s, or Gigabit speeds.
Data is conveyed using fast link pulses (FLPs), which
are merely sequences of NLPs that form a message.