Please explain why an RJ45 splitter won't work

it's not like an extension cord... it's an intelligent electronic device.

The bits are formed into an Ethernet frame and transmitted on the media. The receiver gathers the bits, buffers them, and re-creates the full frame. The hub acts as a go-between to perform the same functions.

If you connect 2 computers together on the same wire at the same time, as you are attempting - then what happens when both transmit at exactly the same time, or both receive at exactly the same time...... the bits would be all shuffled together creating garbage.

That's why the Ethernet standard for twisted pair wires requires a hub in the middle. The original coax standard, using T-'s - did exactly that... share the media. However, the network card would first "listen" to see if anyone else was transmitting, and if not - then it would transmit. If collisions happened, then each station would back off, and after waiting a random amount of sub-second time - try to transmit again. With the Ethernet twisted pair standard, this has changed somewhat.

Hubs are powered, electronic devices that perform all the magic - they are dumb in one sense - but they are intelligent... not just splicing the wires together - else they would not be powered.

If you need more - do a search on Ethernet wiring -

Tomás Ó hÉilidhe wrote:

yes - look up -> Ethernet CSMA/CD

Tomás Ó hÉilidhe wrote in part:

Not simultaneously. Even plugged in (with or without second patchcord) it may introduce enough discontinuity to disrupt signalsi by resonance.

OK, I'll try: learn transmission line theory. It's very complex, but the basics are easy: A signal (current/voltage) can be put in a carefully designed line and be correctly detected at the far end. One important element of transmission lines is termination, usually something like a 50 ohm resistor to ground to limit resonance and give predictable measurement impedence.

In older systems (10base5, 10base2) these resistors were external and the daisy-chains troublesome. 10baseT, 100baseTX and now 1000baseT (gigE) are star-wired with the terminating resistors built-in. So if you go "Y", you've just double-terminated that line.

Worse, the "Y" instroduces a split for the signal. Its' strength will be dropped in half and the reflections may set up resonances. You might be lucky and it works, but no guarantees of any kind. Just like you cannot splice

10/100/gigE with full reliability.

I look upon modern ethernets as something of a miracle: Ethernet ran 100 MHz over 100m of wild country at a time when motherboard designers couldn't get 50 MHz reliable over 10cm of perfectly controlled PCB. Balanced signalling is a big part of the answer, but clean lines also matter.

Yes, it receives, regenerates (cleans up), amplifies and retransmits signals. Fairly simple functionality not requiring any memory or "intelligence".

The hub does do processing.

The simple answer to your question is that the types of Ethernet that use RJ-45 plugs ARE NOT the types that allow paralleling of hosts along a single bus. If they were, then your could do this sort of splitting.

The flavors of Ethernet that use RJ-45 plugs depend on star topologies. Only two devices connected on each link, one on each end. Most of these have separate transmit and receive twisted pairs, for one thing, which means that simply paralleling the Tx and Rx pairs will not work.

Think about it. If you have multiple hosts sharing a single RJ-45 connection, with paralleled Tx and Rx, how will one host's transmission reach another host's receiver? Is there any other way to connect multiple hosts in parallel, when separate Trx and Rx twisted pairs are in use?

In principle, you could use a single twisted pair, and connect all hosts in parallel, using the same protocol as 10BASE2. Although this is not easy with twisted pair, due to the less than reliable impedance characteristics of twisted pair (which creates any number of reflections along the line), Rich Seifert did explain on here some time ago that Tut Systems developed just such a system. It was developed for layering Ethernet over common household telephone cabling. Analog telephones in fact are often paralleled that way. But note: only one twisted pair was needed for the bus. This avoids the problem of deciding how to parallel separate Tx and Rx twisted pairs. And also, the distances involved are typically very short. Combined with relatively low bit rate, which means that symbols along the transmission line are not tightly packed, this can work well.

Bert

P.Schuman:

I was under the impression that all network hosts perform CSMA/CD. Are you saying that only hubs do? (I'd like to leave switches out of the conversation for now)

When you connect two computers together with a cross-over cable, is the reason you don't get collisions = that you have separate pairs for sending and receiving?

When two computers are connected with a cross-over cable, you can indeed get collisions. The point is, both computers can detect the collision, because the Tx pair of one goes to the Rx pair of the other. A computer would detect the collision when it sees energy in its Rx pair, while at the same time it is transmitting. So with two computers set up this way, no problem.

Now parallel multiple computers on a single RJ-45 connection. How do you do that? All Tx spliced together, all Rx spliced together?

Okay. let's try that. One of the computers starts transmitting. How do the others hear that transmission? Certainly not from the signal present on their Tx pairs, right? So how?

Bert

all Ethernet RJ45 network devices - the network card, the hub port, the switch port, the router port, etc

It is part of the IEEE 802.xx specifications for how Ethernet works.

YES

Albert Manfredi wrote in part:

This is not a collision because no data gets mangled. This is just normal full duplex operation.

-- Robert

There is nothing to PRECLUDE setting up two computers interconnected by a cross-over cable to half duplex mode. In which case, collisions can certainly occur.

Bert

what are you guys reading to learn Ethernet ????

When a transmit happens, the bits go to the media, and the voltage waveform increases. For a collision to happen, more than one device on the same transmit line has to apply bits, which increases the voltage above the norm, and that's how a collision is detected.

SO - if there is only 2 computers connected with a crossover, then where is the additional ghost computer "sending" on the same transmit wire ?

Or think of it this way. If multiple computers are connected via and L2 switch, no data would ever "get mangled" either. And yet, if these compouters are set up for half duplex operation with the switch, collisions can occur.

When separate Tx and Rx pairs are uised in an Ethernet interface, half duplex operation may simply mean that simultaneous use of the two twisted pairs is not allowed. Even in cases where data would never get mangled.

Bert

You really have to get beyond 10BASE2. What you explain here does not apply to 10BASE-T or anything else beyond 10BASE2.

If two computers are connected with a crossover cable, is there anything to prevent both from beginning to transmit RIGHT NOW? Certainly not.

If these two are set up to *half duplex* operation, will the simultaneous energy in the two twisted pairs be permitted? Certainly not.

Of course, they can also be set to work in full duplex.

Now get the the original poster's question. Tell me how you can simple parallel many computers on a single RJ-45 plug. The reasons given previuously why this won't work did not even address the most obvious problems. Issues with impedance, for example, are secondary.

Bert

Only after a fashion. If the etherchip waits for silence and starts transmitting and receives a signal during transmission it won't produce any physical collison on the crossover. But it will detect what other stations might receive as a collision and send out a jam signal since it cannot know what is on the other end of the wire.

-- Robert

But that's exactly the case with any Ethernet collision, once we get beyond use of coax or simple L1 hubs. As long as Ethernets out there continue to use CSMA/CD in sysrtems with L2 switches, for example, any collisions are manifested the same way as they are in the case of two computers with crossover cable (set to half duplex).

Bert

- a little light reading may be in order -

Good.

Here is what your wrote:

Be sure to point out to me, in the reference, where multiple computers transmitting simultaneously, using (for example) 10BASE-T, increase any voltage or current in any cable in the network.

Bert

No. If the L2 switch and etherchip are set to or auto-neg full-duplex, then that's what it is and collisions (data damage) cannot occur on that segment.

However, if other segments on the L2 switch are half-duplex, then the switch might receive jam signals on those segments. It should just do the backoff thing on that segment alone, but who knows if cheaper switches isolate that effectively.

-- Robert

I'm not disagreeing with you. I'm just trying to point out that when L2 switches are used in half duplex, which is very often (still) how they are set, there is never any mangling of data possible. And exactly the same applies to two PCs interconnected by a cross-over cable, and set to half duplex.

Mangling of data is never the issue. But collision CAN occur. All "collision" means in these examples is that there is simultaneous energy over both twisted pairs. The fact that this might not cause any "data mangling" anyway, and it certainly does not in systems with L2 switches, is not the issue.

That's all I'm trying to get across.

Bert

Albert Manfredi wrote in part:

This is an unconventional definition of "collision". Cat3 is designed to handle the EMF crosstalk at 10 MHz, Cat5 at

100MHz, and Cat5e from all 4 pairs at 100. Collisions are on the same wire (or logical net through a hub), not crosstalk.

-- Robert