Splitting an Ethernet cable to connect two clients is not as impossible as some answers here may suggest. Nor it is unheard of.
The main issues you will introduce by wiring your network like this is:
- the clients can't make use of the CSMA/CD collision detection and thus making them
- the clients can't talk to each other
1.1 Clients can't talk to each other issue
Both clients can talk to the router just fine (when we ignore the collisions) but they can't directly talk to each other because one client's transmitter wire is connected to the other client's transmitter pin. The correct wiring would be transmitter-to-receiver and vice versa.
I didn't dive into this particular problem too deep, but it seems possible that the router would traverse the communication between the two clients, so you really just need to solve the collision problem.
On the other hand the router may discard traffic addressed for endpoint that is physically on the same router port as origin. The router may rightfully expect hub downstream.
1.2 Collision issue
This is really just consequence of previous problem. The clients and the router listen on their own receiver pin before and during a transmission. If they detect somebody's else transmission they postpone or interrupt own transmission. This and how to resolve collision describes CSMA/CD.
So router can use the collision detection, but the clients can't. This will mumble some or all clients traffic.
There is one important point to mention - the CSMA/CD listens on receiver pin and it does not and cannot listen on transmission pin. One could argue that both clients could prevent collisions by detecting traffic on the other client's transmission wire which seems sensible as that is how we split the cable. This would be holy grail for us as it would let us split the cable as we like, with no collisions. However that is not the case because:
- your own transmission would trigger false positive collision detection
- you need to detect collisions even when you currently transmit
Let's discuss some options.
2.1 No-effort solution
Just splitting the wire and see what happens. It could work as long as there are no or little collisions maybe because:
- the collisions are not that bad (i.e. yes the network is flaky but it works)
- the clients are not that chatty (causing little collisions)
- or they receive only (e.g. Wireshark wire tap)
Collision is dirty word, but actually I have no idea how bad they are on real world network.
2.2 Solving collisions by other means
- the two clients are never powered at the same time
- (few more ideas came across my mind, but nothing practical or interesting)
If the collisions are too bad and you can't utilize the built in CSMA/CD you are pretty much screwed.
2.3 Using Ethernet splitter MYWA-04, MYWA-08
This is not real solution, rather workaround. Those splitters sacrifice 1 Gbps speeds in favor of two 100 Mbps independent ethernet channels in one wire. It comes with some issues discussed elsewhere, but I list it as option.
2.4 Solving collisions by on-wire hub
Hub is the solution to your problem. Its main function is to resend incoming traffic to all other ports except the traffic's origin (which would trigger false positive collision detection). That's it and it also summarize the problem we dealing with.
You could use couple of diodes to clone one clients transmission to the other client's receiver and vice versa. That would create simple unpowered passive hub.
It would be cool to modify MYWA-07 for this:
2.5 Three-port unpowered passive hub
This is similar to previous idea, but for 3 clients. Kudos to Miroslav Adzic.
The internals are explained in Building a passive ethernet hub with anti-parallel diodes
2.6 PoE hub or switch
One can also consider using PoE hub or switch if for some reason you are constrained on the other side of wall (so to speak).
- this all assumes half duplex mode, which really is synonym to communication over shared media (e.g. walkie talkie)