1-2
Switches on the other hand, are intelligent devices used to more efficiently connect distributed
Ethernet nodes than a hub. Unlike a simple hub, a switch provides
targeted
data transfer, as it
will forward a data packet to a specific port or network segment, rather than to all ports, thus
freeing up bandwidth. The ability to target a packet to a specific port increases network
throughput and helps to eliminate the collisions that have historically made Ethernet
non-deterministic.
So we see that by targeting the data transfer between ports, switches act as intelligent repeaters
to increase network distances. In doing this, they actually split networks into separate collision
domains at each port. Thus, switches increase determinism by reducing collisions. Switches also
increase network bandwidth and throughput, as well as provide supplemental error checking on
data packets to help ensure the integrity of forwarded data. Each port of this switch functions just
like any other Ethernet device. It is able to receive and decode Ethernet frames, test for frame
integrity, plus assemble and transmit Ethernet frames.
With Ethernet, any device can try to send a data frame out at any time. If two devices happen to
send a data frame at the same time, then a collision may occur. The arbitration protocol for
carrier transmission access of the Ethernet network is called Carrier Sense Multiple Access with
Collision Detect (CSMA/CD). With CSMA/CD, each device will first sense whether the line is idle
and available for use. If it is, the device will begin to transmit its first frame. If another device also
tries to send a frame at the same time, then a collision occurs and both frames are discarded.
Each device then waits a random amount of time and retries its transmission until it is
successfully sent.
Unlike other Ethernet devices, such as an Ethernet host adapter or Network Interface Card (NIC),
the port of a switch does not require its own MAC address. During retransmission of a received
packet, the switch port will instead look like the originating device by having assumed its source
address. This is why the Ethernet collision domain is said to terminate at the switch port. That is,
a two-port switch will effectively break a network into two distinct data links or segments. An
eight port switch like the Acromag EIS-408FX can break a network into 8 distinct data links or
segments (also called
collision domains
). Since all Ethernet nodes are able to recognize the
occurrence of a collision, and since the detection of a collision is principal to the way Ethernet
arbitrates media access, large domains containing many nodes can become quite cumbersome.
Thus, using an Ethernet switch to subdivide a large network into separate collision domains will
certainly help to increase throughput.
The current tendency in critical industrial control applications is to connect one Ethernet device
per switch port. This will produce the most deterministic mode of operation as the switch can
then operate full-duplex, with no chance of collisions. This ensures determinism, helping critical
control applications to remain predictable and on-time.
Each port of a switch forwards data to another port based on the MAC address contained in the
received data packet/frame. In order to know which port to forward a data packet to, the switch
will learn and store the MAC addresses of every device it is connected to, along with the
associated port number. However, until the switch actually learns the port a particular address
resides at (the first packet), it forwards this traffic to all ports just like a hub. The switch will its
internal look-up table to quickly determine the location (port) of a node, establish a temporary
connection between itself and the node, then terminate the connection once a packet is
transferred. In this way, it increases network bandwidth and provides the network determinism
required for critical control applications
.
This switch uses a
store and forward
algorithm to process Ethernet frames. That is, it first stores
the Ethernet frame and examines it for errors before forwarding it to its destination. Although this
method may seem to increase the forwarding time (latency) and possibly cause fragmentation, it
effectively reduces the occurrence of error frames and improves overall throughput. This is
particularly useful when there is heavy network traffic and or greater potential for noise and
interference.
