Why do we have both MDI and MDI-X?

Question

I understand which types of devices tend to use MDI or MDI-X, but why don't all devices just use one of them? It seems like Auto MDI-X was the solution to this but it still seems like all newly manufactured devices could just switch over to one. This way only crossover cables would be needed for all related connections.

Answer 1

The ability to configure the logic into a device to sense whether a pair is being used for RX or TX in networking is relatively recent and until included (as an option) in Gig Ethernet, was relatively expensive.

So before this time, a pair was designated at TX on one side and RX on the other. For proper communications, a "cross" was required at some point in the connection to make this work.

So where best to make this "cross?" While considering this, one needs to keep in mind two "crosses" will negate each other (or that you need to maintain an "odd number of crosses"). Here are the options:

1. Use crossover cables to make connections
2. Cross the infrastucture cabling
3. Manufacture a cross into one of the two endpoints

In the first case, this works fine if the two devices are directly connected. But if infrastructure cable were used, then you would need a crossed cable on one side and a straight cable on the other. Or, while less than ideal itslef, one runs into situations where two cables are "coupled" together and this would require a straing/crossed combo to work. It is much simpler to manage and less prone to human error if you were to use straight cables on each side. So this is less than ideal.

In the second case, this can have several problems, but most can be avoided by combining with the first case and use crosses with both cable and infrastructure. However the problem comes in when you go to manage the cable infrastructure itself. Namely, both sides (near and far) will use different patterns. Which is used on which side? What if this were mixed over time (maybe by different staff with different personal "standards")? What about the case where you have three (or more) rooms all interconnected by infrastructure cabling? Again, this can be more complex to manage/troubleshoot as well as prone to human error.

The third case actually provides the simplest and most easily standardized way to introduce a cross. If the L2 network device always provided a crossed connection (MDI-X), and the end user device is always straight through (MDI), then it makes the "rules" very simple. You always use a straight connection (cable and infrastructure) unless you are connecting two end user devices (MDI to MDI) or two network devices (MDI-X to MDI-X).

Since this will cover the majority of connections/situations, this makes it less prone to error and allows an entity to maintain most of their inventory as straight cables with no or few crossed cables (as they can typically be ordered at the time of need - with the network equipment). No one needs to remember which side of the infrastructure cabling needs to use which pattern.

Answer 2

This is a matter of DTE vs. DCE -- the role each endpoint takes. The NIC in your desktop is a "user terminal" (DTE). The switch it connects to is "communication equipment" (DCE). All this means is which pins are transmit vs. receive, such that a straight through cable from DTE to DCE works as expected. This is why a crossover cable is(was) required to connect two switches (DCE) or two hosts (DTE) together.

"Auto" really didn't become popular (read: everywhere) until the era of gigabit ethernet. Gig-e uses all four pairs for both TX and RX, so the required logic for auto-mdix is already there. Prior to this, additional logic (read: additional costs) were required.

Answer 3

This is one of those design decisions, which in retrospect seems like an obviously poor choice. But once the problem with the design decision became apparent, the installed base was already too large to simply undo the mistake.

One might think this could be fixed as part of standardization on a higher speed, which has happened several times since the problem became known. But since crossover cables are a minority, it is not that simple.

If we wanted to fix it, we would have to proceed in multiple stages:

• Make auto MDI/MDI-X negotiation part of the standard as upgrading the speed
• Make cross over be a requirement for cables
• Make auto MDI/MDI-X be an optional part of the standard

At each stage we would have to wait for enough old equipment to be replaced, such that we wouldn't cause a problem by taking the next step.

But the problem is even harder to fix. Because 10BASE-T and 100BASE-TX crossover cables only cross over two of the pairs, while the other two pairs are still straight. So most of the crossover cables that we already have are insufficient to completely eliminate the auto negotiation. So at the first step, we would have to standardize on auto negotiation with support for three different kinds of cabling, straight and two different kinds of crossover cables.

But the part about only three ways to wire a cable being required is an oversimplification. If you start mixing the different kinds when patch cables are used at each end of a link, there will be four different ways the cabling could come out. And that is assuming each of the parts of the cable run were done correctly.

Moreover if a new standard mandated MDI/MDI-X/.../... auto negotiation with support for all four ways of crossing or not, what are the chances cable manufacturers would start manufacturing only such a new kind of crossover cable? After all, the old cables without crossover would work with all devices, the new standard cables would only work with the newest devices.

Additionally one has to remember that crossover must only happen on the cables that connect male to male or female to female connectors. If you have any cable run with a male connector in one end and a female connector in the other end, that particular cable need to be straight. Failing to follow this requirement on all your cables would mean they could be connected with an even number of crosses on the path, which would result in an uncrossed connection between the two devices.

In particular any female-female connector you already have for connecting two patch cables together could be incompatible with the all crossed approach if that female-female connector did not also cross the connections.

With all that complexity involved in retroactively fixing a poor design choice, it seems obvious why it isn't going to happen. It is not even obvious, that it would be desirable, even if we could eliminate all the old straight cables.

If we imagine a world where all cables were crossover, and MDI/MDI-X negotiation was an non-existing concept, straight cables would occasionally happen by mistake when connectors are manually installed on cables. So auto negotiation might actually be the desirable design, in particular since we already paid for it being developed and standardized.