Hubs and switches.
A hub is really nothing more than a wire directly connected to each and every port. What
goes in one port is sent out all ports at the time of transmission. If two or more nodes (a
node is anything capable of sending data on the network) transmit at the same time a
collision occurs -the data collides on the wire, becomes scrambled, and has to be
retransmitted later. The entire hub is called a collision domain* because any computer on
any port can collide with any other computer on any other port. Think of this a party line
or conference call. Everyone else can hear your conversation and must wait for you to
stop talking before they can speak. If two or more people talk at the same time (a
collision) the conversation has to be repeated.
*Note: A collision domain is any wire that can have 2 or more devices trying to
transmit at the same time, rendering all transmissions useless. After detecting
the collision, all units must reset and attempt to re-transmit the data. The
more traffic on a wire, the more likely collisions are to occur. The more
collisions, the more retransmits which means more traffic on the wire. (See the
problem?) Ultimately, the network gets REAL sluggish.
Switches are a vast improvement over hubs because they limit Collision Domains.
Switches move packets across the wire from port to port. They handle this task in
different ways depending on how the switch is configured. One way, called "store and
forward", will wait for an entire valid Ethernet frame before moving the packet to the
destination port. This was is very safe since you reduce the passing of bad frames nearly
completely. However, "store and forward" is also the slowest switching method. "Cut
through" switching will look at the packet coming in. It will look only at the beginning of
the packet where the source and destination Media Access Control (MAC) address is
kept. After that, it "cuts the packet through as it comes in", straight to it's destination port.
This form of switching is extremely fast since the switch never has to wait and examine
each Ethernet frame as it comes in. However, because the switch DOES NOT do this,
there is a chance it may pass a bad frame. So there are higher bad frame rates on switches
configured for "cut through" switching. Most common switches will allow both of these
formats to be assigned, and some even allow hybrids of these two methods. One such
hybrid is "Fragment Free" switching mode. The smallest legal Ethernet frame is 58 bytes.
Fragment free switching gets the first 58 bytes and then cuts the packet through to the
appropriate port. Performance is increased by not requiring a switch receive the entire
packet before forwarding it, and yet still cut down on the amount of network errors that
pass from one collision domain to another.
Switches store and forward or cut through packets on a per port basis. If the computers
on port one and two are transferring data, those packets are not replicated on every port -
freeing up the computers on all other ports to pass data without waiting for useless traffic
to get off the line. Now, if there is just one computer on each port, it cannot collide with
any traffic. This actually opens up the possibility of full duplex operation. If the computer
and the switch are the only things that will ever talk on the wire, then they never have to
listen for collisions because their transmit signals are hardwired to each others receive
Comentários a estes Manuais