Few questions on full duplex/ half duplex.

Please correct any incorrect assumptions. Thanks

1. Both 10base T and 100BaseT can run in either half duplex or full duplex.
When 10baseT operates in half duplex, it uses how many pairs? 2 pair?  What about when it operates in full duplex?  Does it use 2 pair still?  The speed of 10baseT half duplex is 10mbps.  But does this speed increase if 10baseT is run in full duplex?

2. Now 100baseT operates at 100mbps in half duplex.  And it uses all 4 physical wires?  100baseT full duplex uses 2 pair?  What is the speed of 100baseT full duplex ....200mbps?

3. Switches provide point to point connections for hosts, so they can communicate in full duplex. What happens if your network card is only capable of half duplex. Does the switch then act like a "hub"?

4. Hubs can only run in half duplex because they dont provide point to point connections. They forward all frames they receive to every connected host.

If you have a 6 port 10/100 hub, and all clients have 100mb nics.  The 100mb is spread out over all 6 users.
If you have a 6 port 100mb switch, and all clients have 100mb nics. Each user gets 100mb.

5. What is the point of having a 1gig backbone, if all of your desktops only have 100mb nics?  

Who is Participating?
lrmooreConnect With a Mentor Commented:
1. Yes, 10Mb can run full duplex. It makes no difference if using full or half duplex, ETHERNET specs requires use of 2 pairs. Speed does not increase, but throughput does - up to theoretical 20Mb because it can send and receive at the same time and not have to worry about collision detection.

2. 10 or 100, does not matter. The same 2 pairs are used. Full duplex is theoretical 200Mb throughput, yes.

3. No. Only that one port will operate in half-duplex mode. That means that one port must enable collision detection and collision avoidance mechanisms, but it does not turn the whole switch into a hub.

4. True. A hub is shared media between all hosts. All packets are forwarded to all ports always, even collision notifications.
A 10/100 "switching hub" acts a little different. It's more of a hybrid. It creates two different "hubs" that are then "switched" between them. All devices connected at 10Mb are in a 10Mb hub, and all devices connected at 100Mb are in a 100Mb hub, and the two hubs are switched. The 10M hosts are all in the same collision domain, and all the 100m hosts are in the same collision domain. The 10M hosts and the 100M hosts are not in the same collision domain, but remain in the same broadcast domain. So yes, if all clients connect at 100m, then all ports are in one shared 100m hub. Each client gets 1/6th of 100Mb. However, remember that in shared media collisions are a fact of life. The higher the traffic, the higher the collisions. The higher the collisions, the lower the throughput. General rule of thumb is that you get maximum of 60% throughput, or 60M devided by 6 - 10M per client.. Contrast that with a switch, running full duplex, each client gets full 200Mb regardless of how many others are connected.

5. If  you have 6 clients all capable of running 200Mb on one switch, and a server attached to another switch, and the two switches are connected with only a 100Mb uplink, then all 6 clients have to share that uplink and that creates a bottleneck.
If the one and only server is only connected at 100M, the it can't service all clients any faster than that anyway, so yeah, what's the point of the Gig uplink? Because we CAN! If we build the foundation and the infrastructure to be the fastest possible, then all you have to do is change out the NIC on that server to a Gig nic and now the uplink in the network is again the bottleneck.
dissolvedAuthor Commented:
lrmoore: Thanks for the prompt response.

4. So you're saying there is such thing as a "Switching hub."  And depending on the speed of clients, it groups them accordingly. So if we have a mix of 10mb clients and 100mb clients plugged into a "switching hub", then the switching hub creates two collision domains. One is for the 10mb users, the other is for the 100mb users.  So essentially it becomes a 2 port switch? Where are "switching hubs" usually found? Is any switch that is 10/100 considered a switching hub?

5. So they key is to have your uplinks operate at greater speeds than your clients. But cant two switches connected via cross over (aka uplink) run at 200mbps if it's full duplex anyway? Meaning all clients and uplinks between switches will all be 200mbps?  Confused on this part

Thanks for the great explanation on the rest!
4. Switching hubs are/were a short-lived hybrid. http://www.linksys.com/products/product.asp?grid=34&scid=31&prid=148
Anything that says 10/100 "HUB" is one of these hybrids.
Anything that says 10/100 "SWITCH" is a true switch where each and every port is a true collision domain in itself

5. The link itself is full-duplex, but that is not full duplex communication end-to-end between the client and the server.
The 200Mb is "potential" because you get 100Mb each way simultaneously. You can't "push" or "pull" at 200Mb. You can send one file and receive another file at the same time for an aggregate throughput..but remember the full conversations of TCP/IP requries acknowledgements for every packet, so it's more statistical than real.

HTH..   <8-}
The new generation of project management tools

With monday.com’s project management tool, you can see what everyone on your team is working in a single glance. Its intuitive dashboards are customizable, so you can create systems that work for you.

PennGwynConnect With a Mentor Commented:
Half duplex on a link means that when either side transmits, it uses the entire link.  Full duplex means that when transmission is happening in one direction, other traffic can still use the other direction.  Some links are inherently half-duplex because there is only a single physical medium for traffic in either direction.

1. 10 and 100 Mbps Ethernet use two of the four available pairs, one to send and one to receive.  If collisions can occur, the sender uses the receive pair to detect them, so it's not available for other traffic.

Half duplex provides 10 or 100 Mbps of bandwidth, which may be split arbitrarily between sending and receiving.  Full duplex provides 10 or 100 Mbps of bandwidth to send, and another 10 or 100 to receive.  If close to 100% of the actual traffic is in only one direction, full duplex will not speed it up.

2. 10 and 100 each use two pairs, or four physical wires, of the 4 pair/8 wires in the cable.

3. Switches CAN provide point to point connections for hosts, so they can communicate in full duplex. But most switches can also operate in half duplex, either explicitly configured or negotiated with the client devices.  You must do this if you're hanging a hub off of a switch port.
> Does the switch then act like a "hub"?
No.  A hub distributes incoming packets to every other port.  A switch will only do that if it cannot determine what port the packet is intended for.

4. Hubs can only run in half duplex because they dont provide SEPARATE COLLISION DOMAINS.  They have to support collision detection, which means that only one sender can talk at a time.

> So essentially it becomes a 2 port switch?
No, there is a 2-port switch (also called a "bridge") inside the box.

> Is any switch that is 10/100 considered a switching hub?
No.  Switches/bridges introduce latency because they buffer traffic between one collision domain and another.  Hubs generally allow traffic to flow directly -- it's repeated, but not buffered.  A switching hub adds the buffering of a 2-port switch to accomodate traffic between ports at different speeds.

5. The backbone can be carrying traffic for all clients at the same time -- but only if it has the aggregate bandwidth available to do so.

> Meaning all clients and uplinks between switches will all be 200mbps?  Confused on this part

The clients each have their own collision domain at that speed.  But the uplink is *shared* by all of their traffic; if a single client can fill the uplink, it becomes a bottleneck.

dissolvedAuthor Commented:
Got ya. So the uplink is shared by all the traffic. A single client can allocate 100mb of bandwidth and thus create a bottleneck. Is it the general rule of thumb, to make the uplink always faster than your clients to prevent bottleneck?
That's a good rule, but it's not always necessary.  For instance, I have a switch here that's talking gig to the rest of the network, and has a 12-port gig card for local connections.  But the backplane only supports 2.4 gig to the card at any one time, so the odds are that any situation that overloads the uplink is also going to max out the backplane -- adding uplink capacity won't really do much for me in this case.  On the other hand, although the local ports are gig, I don't expect any of them to try to stream data at gig speeds in the ordinary course of an average day.

wmilligaConnect With a Mentor Commented:
Not to beat this one to death, but in general- it you have a 10baseT device, it usually implies a hub and will not typically support full duplex. When a device supports 10/100, it with typically run full or half duplex at either speed. In many cases you need to hard set the duplex on both ends (host and switch port) to avoid duplex negotiation errors- the port will still run if both ends do not negotiate properly, but will be painfully slow due to errors.
dissolvedAuthor Commented:
Ok one last time.

Half duplex. Uses 4 wires.

2 of the wires Rx
2 of the wires Tx

But they can only do one or the other, not both at the same time

Full duplex. Uses 4 wires as well.

2 of the wires Rx
2 of the wires Tx
at the same time.
dissolvedAuthor Commented:
pengwyn, what is a backplane?
All Courses

From novice to tech pro — start learning today.