Need help with port duplex settings

Which is exactly what the standards say should happen. Not that this is what anybody expects, but it is what it is.

When you disable auto-speed, you also disable auto MDI-MDI/X detection. Whatever cable you are using probably is the wrong one.

Perhaps the upstream provider handed you a port expecting a straight through cable to save you future problems, and then you went and put a crossover cable there anyway...

Yes, turned out to be the wrong cable. We've done many, many similar sites in the past and have always used a crossover cable, which I have installed. Someone else did it on this particular site, and I assumed (too much) that it was a crossover.

It always surprises me that people seem to think forcing one end will make the other follow suit.

Negotiation has improved hugely, and now normally just works. Add i the auto MDI and I actually prefer to let switches sort it these days.

Most of the negativity comes from te early days, where with a 3Com lan card and a 3Com switch, you had an evens chance of negotiation being successful. Those days are long gone.

Paul.

If you think of the setting as being "this is what I want - please do the same" then it makes sense. It's the fact that it's "this is what I'm doing and I'm not going to talk to you any more" that seems to be unexpected.

We had different models of 3Com switch which resolutely refused to negotiate a working link. That was long ago and I think the switches concerned had come from different companies that 3Com had absorbed.

Sam

But what is important is: it is all a thing of the past. It is not a concern anymore.

But still, there are companies that keep this "thou shall set the ethernet port to a fixed configuration and not use auto" in their standard practices, thus causing extra maintenance effort and non-working setups, without any advantage.

IMHO setting anything auto is proof that you don?t understand networking properly.

I have had numerous weird faults mainly with Video and Voip which have been traced to this problem .I always lock things to some fixed setting or other

If you force the switch port to 100 full then you can be sure that everything that can connect to them must work at that setting and all the cables are correct. And they are getting the bandwidth they are paying for.

As a starting point cable wise ,any ports of the same type of device IE switch router etc think crossover different device type think straight .It is not an absolutely foolproof method but it is a start.

Tip here ,take a short length crossover cable with you at all times and a cat5 straight through coupler. If you disconnect the cable and add it you change to the other sort of cable . Add it to a straight and it becomes a Xover and an Xover to an Xover becomes a straight.

Oh, I agree, I run thousands of ports in auto with no problems. Handoff to customers, handoff to my servers all work with no issue.

But MOE type handoffs by the phone companies are typically all set to hard code for speed/duplex for sub-gigabit speeds. That is their standard and that is how it is.

I never had any problems with 3com, but there was one specific cases with Sun SPARC servers and Cisco switches that came up every now-and-then. But last time I ever saw problems like that was back in the mid '90s. By the late '90s, everything worked all the time for autonegotiation.

But that doesn't change the ingrained ideas that you *have* to force duplex/speed on 100Mbps ports in many circles.. Even as recently as the Cisco PIX era, the docs strongly recomended hard coding duplex/speed, so usually whenever people brought in a PIX, I'd have to match them on hardcoding duplex/speed.

FWIW: The GigE 1000-Base-T standard doesn't require auto MDI/MDI-X, but it is suggested that it is supported.

There are a tiny few 1000-Base-T things that don't support auto MDI/MDI-X, but not many. They would need a crossover going switch-to-switch.

Isn't that "enable only a single negitiation outcome" in the auto negotiation setup? I think auto negotiation is mandatory in GigE, only you can configure a set of acceptable outcomes at either end.

Setting this to "1000 fulldup only" just means that the link will fail whenever this cannot be negotiated with the other end.

Our existing telco has that "100 fulldup hardcoded" idiocy in its glass termination units as well. Fortunately the new one that will bring in the equipment this month asked us what we want, and I specified auto negotiate. (and I have to remember to remove that hardcoded config from the port before it causes trouble again)

I used to work for a telco that does just that , it prevents customers kit auto negotiating to 10 half. Then they call in some weeks later with a fault saying we cant get any more throughput than 4meg ,that is the reason for hard coding ports clueless customers. The ones that know what they are doing don?t have a problem.

We always did speed test before handover ,but that was to our test gear not the customer kit. Not my decision.

Hot desk area?

I assume you use static routes everywhere rather than those new fangled routing protocols?

Hot desk area?

Does the desk temp affect the lan connection ? works just fine on my desk but that is at room temp.

No BGP mostly and OSPF for the interior protocol

Do you have a reference to which data bits are sent down which pair in

1000BASE-T? Since all four pairs are used in both directions simultaneously there seems to be no point in using anything other than a straight-through cable in any situation, though I can see the point in detecting a crossover and adapting to it.

And are the tiny few the some ones that don't support any other negotiation either, such as very old Cisco 12000 GigE cards?

Sam

That's not our experience. GigE requires that if negotiation fails then the port be disabled, not that it falls back to some lowest common capability, i.e. HD. The result is that you get one end (the non-negotiating one) setting the link up and the other insisting it's down. You can spend a lot of time messing with cables before you realise what's going on.

Again, negotiation is either on or off. Hardcoding settings means that negotiation is off, not that the system tries to negotiate those settings.

Yay!!

Sam

Surely it ensures that customers' kit falls back to HD rather than preventing anything.

You need a clueful customer to spot that they need to turn off autoneg and hard code the speed and duplex.

So you can tell the customer "it's working fine, there must be a problem with your kit"? :-)

Sam

When I understand it correctly, there is no "autonegotiate off" with gigabit ethernet. Autonegotiation is always on. But you can configure what results you consider acceptable outcomes of the negotiation. When you configure it to allow only 1000 full, you are not turning autonegotiation off, but you are telling it to autonegotiate and to fail if it cannot do 1000 full (e.g. with a 2-pair cable).

I suspect you can getting mixed up between what the standard says everybody should do, and what some manufacturers do anyway.

To be fair some early GigE ports sometimes struggled with auto negotiate before standard chipsets were used for everything, and so the dreaded backward compatibility got into the mix.

some kit can still operate without auto negotiation at least on fibre (Cisco and Marconi SDH come to mind).

Some devices refuse point blank to talk to a GigE fibre port without auto negotiation turned on - my favorite was a Foundry switch which would turn on all the lights, but not pass any traffic........

IEEE 802.3ab was the standard for encoding 1000Base-T, although it was superceeded by something else, I didn't follow what.

Its a complex bit scramble method. Some description of it is here.

None of this is related to straight through or cross over though. Its simply that host devices talk one way, and switches expect to receive their talking. If there are two hosts or two switches, you need to cross over all pairs so that they can continue to communicate.

I believe that is one of them, although I only had fiber on the 12k I ran.

The ones I ran into was the early Cisco copper GBICs did not have auto MDI/MDI-X on them.

Aarch, 3Com Vortex cards. Didn't work reliably with Cisco, 3Com or Avaya (1). Funny thing is, I occasionally use such a card nowadays and never have a problem anymore.

M4

(1) In those days Avaya did not work to well with anything, so no surprise there. However, because one of the bugs was with disabling autoneg, and autoneg did not work, we were kind of stuck there....