Duplex Question

Daniel333

It's understanding, that a device can run in full-duplex as long as there is no chance for collisions, right?

But does it have to?

e.g. If I have two switches plugged into each other with a cross-over cat5, I should get full duplex by default. Can I cut that to half duplex?

Theoretical question here, suppose I have two 12' lengths of cat5 and I wanted to interconnect two switches that are 24' apart. Could I use a dumb hub as a coupler in that situation, since it would cross over the cable for me?

as long as nothing else is on the hub, it should still work in full duplex, right? Or am I misunderstanding something here?


thanks as always,

litehedded

Daniel333 wrote:

It's understanding, that a device can run in full-duplex as long as there is no chance for collisions, right?

But does it have to?

e.g. If I have two switches plugged into each other with a cross-over cat5, I should get full duplex by default. Can I cut that to half duplex?

Theoretical question here, suppose I have two 12' lengths of cat5 and I wanted to interconnect two switches that are 24' apart. Could I use a dumb hub as a coupler in that situation, since it would cross over the cable for me?

as long as nothing else is on the hub, it should still work in full duplex, right? Or am I misunderstanding something here?


thanks as always,

should be able to manually set full duplex if it doesn't auto negotiate

Ed Rooney

You will get half-duplex. Switch 1 to hub is a collision domain. Switch 2 to hub is a collision domain. A hub is a single collision domain, therefore the whole mess in one collision domain. Full duplex is only possible with 2 stations in a point to point, like on a switch. Data from switch1 to switch 2 will be pumped right back out through both links.

Take all of the hubs in your enterprise and throw them in the garbage. Hubs cause loops and duplex mismatches which are a cancer to your network.

litehedded

Ed Rooney wrote:

You will get half-duplex. Switch 1 to hub is a collision domain. Switch 2 to hub is a collision domain. A hub is a single collision domain, therefore the whole mess in one collision domain. Full duplex is only possible with 2 stations in a point to point, like on a switch. Data from switch1 to switch 2 will be pumped right back out through both links.

Take all of the hubs in your enterprise and throw them in the garbage. Hubs cause loops and duplex mismatches which are a cancer to your network.

it won't be pumped right back if you disable the loopback circuitry by forcing full duplex in the nic driver right?

Ed Rooney

litehedded wrote:

Ed Rooney wrote:

You will get half-duplex. Switch 1 to hub is a collision domain. Switch 2 to hub is a collision domain. A hub is a single collision domain, therefore the whole mess in one collision domain. Full duplex is only possible with 2 stations in a point to point, like on a switch. Data from switch1 to switch 2 will be pumped right back out through both links.

Take all of the hubs in your enterprise and throw them in the garbage. Hubs cause loops and duplex mismatches which are a cancer to your network.

it won't be pumped right back if you disable the loopback circuitry by forcing full duplex in the nic driver right?

What nic driver? We're dealing with 2 switches connected by a hub.

tech-airman

Daniel333 wrote:

It's understanding, that a device can run in full-duplex as long as there is no chance for collisions, right?

But does it have to?

e.g. If I have two switches plugged into each other with a cross-over cat5, I should get full duplex by default. Can I cut that to half duplex?

Theoretical question here, suppose I have two 12' lengths of cat5 and I wanted to interconnect two switches that are 24' apart. Could I use a dumb hub as a coupler in that situation, since it would cross over the cable for me?

as long as nothing else is on the hub, it should still work in full duplex, right? Or am I misunderstanding something here?


thanks as always,

Daniel333,

I think your question should be how does half duplex work, the problems with half duplex, and how full duplex solves the problems with half duplex.

To understand half duplex Ethernet, imagine you're using a 10Base2 or 10Base5 Ethernet. Since the "networK' part of the 10Base2 or 10Base5 network is the coaxial cable, only one can "speak" on the "network" at a time. The rules for when a network node could "speak" is based on CSMA/CD. For the sake of simplicity, let's imagine two hosts on a 10Base2 network. That's pretty much half duplex in it's purest form. So Host1 "speaks" to Host2 by requesting a file. Well, Host2 cannot reply to Host1 until Host1 "shuts up." Once Host1 "shuts up", then Host2 replies with the file. While the file is being transferred, Host1 now has to "shut up."

Now fast forward to 10BaseT. Imagine now that there's two hosts connected to each other with a crossover cable. Host1 "speaks" to Host2 requesting a file. Once again, Host2 has to "shut up" until Host1 stops talking. Well, with 10Base2 or 10Base5, the way that you know that the other side has "shut up" was that there's no signals being detected on the coax cable. Then any host who wished to transmit could do so. Well with 10BaseT Ethernet, the transmit and receive pairs of cable are separate. The purpose of the crossover cable is to let the "mouth" (transmit pair) on one side of the cable to "speak" directly into the "ear" (receive pair) of the other host. Similarly, the "mouth" (transmit pair) of host2 is connected directly to the "ear" (receive pair) of host1. But the question is, how does a 10BaseT NIC know when the other side has "shut up" in order to be able to transmit? The answer is a loopback circuilt in the NIC from the receive pins to the transmit pins on the same NIC. That way, if signals are being received on the receive pins, then that means the other side is still "talking" and therefore has not "shut up" yet. So when that loopback circuit "hears" silence, then that's how the transmit pins know that it's now safe to "speak."

Now, imagine instead of two hosts connected with a crossover but 12 hosts connected to a 10BaseT hub. Now, 11 hosts have to wait and "shut up" while one is "talking." So one of the problems of half duplex is that many hosts have to wait until the current "speaker" on the network has "shut up" before they can transmit themselves. To help illustrate the problem of the wasted time, imagine you're with a very talkative friend. We're talking a verbal fire hydrant. Let's just say you just passed the CCNA exam so you want to tell your friend. Well, when you meet the friend, your friend says "Hey, how's it going? oh my gosh, did you hear? blah blah blah blah blah blah....." So you now have to be patient and wait to transmit your good news of the CCNA exam pass. Well, your friend just keeps talking and talking and talking so you have to keep on waiting and waiting and waiting. In other words, your time is being wasted. Now finally imagine your friend stops talking. NOW you call say "Hey, I passed the CCNA exam! I'm soooo happy!" Now, probably due to the time delay for you to be able to say what you wanted to say, your happiness of passing may have turned into annoyance. So I hope you now understand the problem of half duplex.

So to solve the problem of half duplex as far as time wasting is to have full duplex communications. Going back to the situation of two hosts connected with a crossover cable, imagine you now have two flexible plastic tubes. One tube is for you to speak into and the other end in your friend's ear. The second tube is for your friend to speak into and the other end in your ear. Let's assume both you and your friend are capable of "full duplex" thinking. So in full duplex, while your friend is talking about what's going on in a soap opera, you can talk to your friend about how to subnet. Since both sides are transmitting and receiving at the same time, it helps to solve the "wasted time" problem with half duplex.

Now to answer your questions:

• Q: It's understanding, that a device can run in full-duplex as long as there is no chance for collisions, right? A: This is assuming either: a) two hosts connected with a crossover cable or b) a switch is the networking device interconnecting the hosts.
• Q: But does it have to? A: No, however I hope I explained the problem associated with half duplex operations.
• Q: Can I cut that to half duplex? A: It depends if you can configure the switch to operate at half duplex instead of full duplex.
• Q: Could I use a dumb hub as a coupler in that situation, since it would cross over the cable for me? A: Maybe. The reason is, overall the two switches need a crossover cable or an effective crossover cable link. The way a hub works is that even though you use straight through cables between the hosts and the hub, the hub itself does some "internal crossovering" for two hosts to communicate with each other. Since the hub does some "internal crossovering", then I guess the switches could communicate with each other.
• Q: as long as nothing else is on the hub, it should still work in full duplex, right? A: Unfortunately no. Once the transmission from Switch1 to Switch2 hits the hub port that is connected to Switch1, the port that Switch2 is in the "wait" and "shut up" mode so the link between the switches would be operating at half duplex.

I hope this helps.

litehedded

Ed Rooney wrote:

litehedded wrote:

Ed Rooney wrote:

You will get half-duplex. Switch 1 to hub is a collision domain. Switch 2 to hub is a collision domain. A hub is a single collision domain, therefore the whole mess in one collision domain. Full duplex is only possible with 2 stations in a point to point, like on a switch. Data from switch1 to switch 2 will be pumped right back out through both links.

Take all of the hubs in your enterprise and throw them in the garbage. Hubs cause loops and duplex mismatches which are a cancer to your network.

it won't be pumped right back if you disable the loopback circuitry by forcing full duplex in the nic driver right?

What nic driver? We're dealing with 2 switches connected by a hub.

d'oh!
I'm an idiot. move along. nothing to see here

georgemc

daniel333 wrote:

Theoretical question here, suppose I have two 12' lengths of cat5 and I wanted to interconnect two switches that are 24' apart. Could I use a dumb hub as a coupler in that situation, since it would cross over the cable for me?

Official Answer: No, you can't and don't want to do this.

Unofficial Answer: If you hard code both of the switches to full-duplex(no auto-detect) and
ensure that nothing else gets plugged into the hub, I believe it will
work. A hub is a dumb device, it won't/can't prevent you from sending
and recieving at the same time. Without a third device
there is nothing to collide with. Adding a third device to the hub will
definitely cause you some heartache though.

Disclaimer: DO NOT attempt this outside of a lab environment. Someone WILL plug a third device (probably more) into the hub. It doesn't matter if it's under lock and key and you're the only one with the key, it WILL happen.

Daniel333

You guys are awesome. You really are.

I don't really work in a real world envionment, and have plenty of extra cat5 and cat6 cable laying around at home, so I won't be actually coupling the switched are stated. Just thinking in theory only.


thanks again!