A sidebar to the article “The Xerox Alto computer,”
published in Byte, issue 9/1981, pp. 62.
The Ethernet network, developed at Xerox Palo Alto Research Center by Bob Metcalfe
and Dave Boggs, is a medium for the transmission of information. It is a
multiaccess broadcast system used to link many computer systems that are physically
within several hundred meters of each other. Each machine in the network is
connected by means of a single, passive, coaxial cable.
The Ethernet network is a packet network. Information to be transmitted from one
node to another is enclosed in an “envelope” of data that describes
the information (ie: its length, a checksum for error control) and describes
the destination of the resulting datagram or packet.
Regulation of access to the Ethernet cable is performed by a system designated
CSMA/CD (Carrier Sense Multiple Access with Collision Detection). This technique works
in exactly the same way that conversation among a group of people at a party
is regulated. A node with something to say waits until no one else is speaking. When
everyone else is quiet, that node begins to broadcast. If another node also
starts broadcasting, this is called a collision. When a collision occurs, all
of the nodes that are broadcasting cease to do so. Each node waits a random
amount of time and then listens for another lull, thus starting the whole process again.
Each node on the network listens to the beginning of each packet. Once a
node has determined that a packet is not addressed to itself, it stops
paying attention to that packet and waits only for the end of the packet to
allow it to prepare to listen to the next one. An Ethernet node may become
“promiscuous” and listen to all packets being broadcast. This allows a
very precise monitoring of statistics on the Ethernet system, since nodes generating
traffic need not perform any statistical processing or use any
network bandwidth to report the statistics. A “silent observer” can
watch the network without ever affecting the performance of that network.
At first glance, the “cocktail party” approach to multiple access
might seem awkward. A second look, however, reveals that the maximum packet
length is only 554 bytes, and the speed of Ethernet transmissions is approximately three
million bits per second. This leads to a maximum transmission time of about
1.5 ms (milliseconds), which means that even on a highly loaded network, the
overall efficiency remains very high.
An interesting aspect is that the Ethernet network itself is entirely passive.
Regulation is done by each individual node. This decentralization means that the
failure of any single node will not significantly affect the network as a
whole. Thus, the mean time between failures of the whole network is very high,
making it a good choice for industrial or business applications where downtime means
The high speed of the Ethernet network rivals that of disk I/O. A file may be
loaded from across the network almost as easily as from a local disk. In a
distributed information processing system, it is highly desirable to allow a file to
be easily accessed anywhere on the network. A highspeed file transfer capability
also permits a new machine to be integrated into the network in just a few minutes.
The competition in local networking is hot and heavy. DEC (Digital Equipment
Corporation), Intel, and Xerox Corporation have recently joined forces to promote
an Ethernet-based industry standard network. Wang Corporation, IBM, and Zilog have
also proposed networking standards. Whatever the outcome, industry seems to have
decided that the future is in local networks.