On Fri, 2 Jul 1999, Murray Armfield wrote:

> Thanks,
> 		That gives me good reason to suspect NT network integration
> with NetBSD. There is a reasonable speed difference between the two systems.
> I will look into it and report back at a later date.
> 	With the comment on Windows drivers, do you mean the network
> settings that Windows detects through the card of the network, or windows
> detection of the device itself.

I'd have to agree with the description that NT is probably not
mis-configuring the card.

> 	Familiar with switches, routers and bridges. Don't think poor hub is
> to blame. Also the hubs at my work are full-duplex as that's the network
> configuration picked up by my network card. Just finished looking through a
> manual for an Intel hub but it didn't explicitly say one way or the other.
> Switches can be configured one way or the other. Bearing this in mind both
> types of hub should be available but I haven't encountered a half-duplex
> only model. A full-duplex model should support half-duplex but probably
> depends on manufacturer.

Ethernet was designed as a broadcast medium. i.e. there's one transmission
medium, and each transmitter can transmit on it. There's a lot in the spec
about timings to make sure that the line is really empty when a
transmitter starts transmitting. Collisions are when a transmitter guessed
wrong.

Since it's a broadcast medium, it is inherently half-duplex. There's no
way around it.

Hubs are just electrical repeaters. They link multiple 10T ports to one
broadcast ethernet.

Bridges break a logical network into two collision domains. A collision
domain is a collection of physically networked machines which can collide
with each other when transmitting.

Bridges have an ethernet interface on each domain (physical network). The
receive each message, and figure out where the recipient is. If the
recipient is on the same side the packet was received on, the bridge does
nothing. If the recipient is on the other side (or it's a broadcast
packet), then the bridge forwards the packet.

The advantage is that two computers talking to each other on one side of
the bridge don't get in the way of computers on the other side.

A switch is a box with a whole bunch of ports, each one of which it breaks
into a collision domain. I.E. a bridge had 2 ports, a switch expands that
to N.

Since a switch figures out what ethernet addresses are on each port, if
there's only one address on a port, only packets to that address will be
sent.

What's happening with full duplex is the cards are: using twisted pair
wiring (so transmit and receive are seperate pins) and assuming that
the link is point-to-point. Thus there is only one transmitter going each
way - it's no longer a broadcast medium.

With a switch, this is fine. It would only be sending info for the one
card at a time down the link. The machine can be sending data at the same
time as there are no other transmitters on its side of the link.

That's the long answer.

The short answer is that hubs (not switches) are standard ethernet with is
half-duplex by defentition. :-)

Take care,

Bill