[PLUG] ethernet switches and glitches

AthlonRob athlonrob at axpr.net
Tue Dec 20 00:28:31 UTC 2005


Aaron Burt wrote:
> With TP ethernet hubs, there were timing issues when you went over X
> number of hubs deep.  AFAIK switches function as re-timing modules and
> take care of this, but each switch has to keep a list of all the MAC
> addresses on each port.  Smaller, consumer-class switches probably
> have a hard time doing this.  Also, they often don't support spanning-
> tree protocol, which is a frame routing protocol for switched networks.

The issue was timing, but not so much a timeout issue.  Hubs are layer 1
network devices.  Switches are layer 2.  Hubs just look at the ones and
zeros, sending them back out to each cable plugged in to them.  Switches
look at the frame's destination and send the frames to the right port.

To avoid network collisions (where two computers attempt to communicate
at once, effectively rendering the information they wanted to send
useless), you could only stack things a few switches deep.  Before
somebody on one end of the network was done sending, there had to be
time for somebody's frame on the other end of the network to start and
get to the first computer.  This way the first computer could jam his
packet, realizing somebody else was using the wire at the same time.
Hubs essentially treated the entire network like a party line - only one
person could talk at a time.  If somebody else started talking, you had
to say "whoops!  I'll try later" and cancel whatever information you
were trying to get out.

With switches, the problem doesn't exist.  Each port on a switch is its
own collision domain.  The switch doesn't just send information it
receives out to everybody, it just sends it to the people who need it.
They're also generally full-duplex, meaning two people can talk at once,
anyway....

Rob



More information about the PLUG mailing list