Can you connect two gigabit fibre NICs together?

Just a quick guess or two: do the cards run single-mode optics while the fiber is multi-mode, or vice versa ? Also, might you need an attenuator inline in case you are overloading the receiver with such a strong signal into it ?

Does anyone recommend a "Fibre for Dummies" Book?

Can someone recommend a "Fibre for Dummies" book? It's not that big a topic at the overview level, really. Make that web site or magazine article.

you need to have the exact part codes for the adaptors - and then check them on the Intel web site.

a quick scan of theweb site implies the adaptors you have are for servers if they have fibre interfaces (assuming they arent a non current model).

there are a lot of versions (there is 3/4 Mbyte PDF describing all the different flavours)

looking at the support pages for drivers

a message about end of support: but this page has a lot of support info, help notes and so on.

a bad idea to look into a fibre port, although you can try shining onto something and looking at that.

AFAIR all the various 1000 Base standards use infrared light?

it should - but check the Intel site in case there is something else needed - or there might be a diags tool of some sort if the card distinguishes between a valid non looped signal and just incoming Ethernet?

They should theoretically work when you connect both Tx-Rx pairs.

Since you can see the dim red light from the TX ports, the cards are using

850nm multimode source. It is considered infrared but the cheaper LED source produces light that is slightly off - both into visible red (less than 850) and into infrared (more than 850), so you are able to see the less than 850nm end of the signal. As long as you are not using any optical magnifier to look at the light and you are at least couple feet away from the connector, you are just fine to look (with a corner of your eye I must add). The aperture of the multimode signal is so large that the light energy spreads right away and you only get a tiny fraction of it when you look (carefully).

Anyways, back to the subject matter: the multimode fiber is susceptible to over-fill condition (too much light), especially on the very short links like yours where the stray modes keep bouncing back and fourth without much attenuation and confusing the receivers. You should make yourself a little mode conditioning mandrel - wrap 5-10 turns of the fiber patch cable around a thick Sharpie or something that's round and appprox 3/4 inch (~20 mm) in diameter. These loops will add the needed attenuation and will strip the outer modes of light off.

Good luck!

(snip)

Visible is usually stated as 400nm to 700nm, but there is a fairly long tail on the long wavelength end. At a high enough power level you can see to 850nm or beyond. Others have stated the best ways to look at such light.

-- glen

Fiber comes in three common flavors. 62.5/125 multimode, 50/125 multimode, and 8.3/125 singlemode.

Yes it should be.

Ok, that's 850nm, so you want to use 62.5/125 graded index multimode cables. Usually, this has an orange jacket.

Incidently, don't look into laser with remaining eye... For eye safety, always reflect it on something. 850nm straight out of the port or with a short cable should be able to shine through a sheet of white paper. Or I perfer to get an IR sensor screen from Lumitek

- disclaimer: I have no financial interest in Lumitek, merely a happy user. I have a model Q-42-R clipped to my ID card so that I can positively identify transmit.

fiber cables need to be properly cleaned - dirt and scratches make a big difference.

make certain the cards show up in dmesg, and that you have started the interfaces.

The card is quite easy to find - it's the oldest PRO/1000 fibre adapter on the site (now discontinued.) Seems to be the only 82542-based card too.

Thanks for the link - I have looked extensively at the Intel site, but I still hadn't come across some of those links. From those you listed I was able to find that the card supports:

- Multi-mode fiber with 50 micron core diameter; maximum length is 550 meters. - Multi-mode fiber with 62.5 micron core diameter; maximum length is 275 meters.

No mention of a minimum length. The cable I have was listed as "MM core

62.5um", so everything looks okay there.

I already tried that by putting my hand in front of the port, but the light isn't bright enough to see. I didn't look directly into the transmitter, but like Dmitri said if you look at an angle and out of the corner of your eye you can see it - of course I only had a quick glance, I figured the lack of warning stickers meant a quick glance should be fairly safe :-)

Haven't checked for diagnostic tools, but "ethtool" under Linux reports plenty of info (the card is quite well supported) but it tells me the card hasn't detected a link.

Thanks for the advice!

Cheers, Adam.

Mine has a black jacket with yellow+blue connectors, and looks like a generic version. But it has "62.5" in the description so for a newbie like me that's enough to assume it's fine :-)

Nope, it won't shine through a sheet of white paper. Can't see it come out of the other end of the cable either, but I don't want to point that straight at my eye to check for certain :-) (as per my other post though, light from a torch makes it through the cable properly.)

The impression I got from the Intel site was that the cards switch to 'low power' mode when there's no link to prevent eye damage, so I suspect this is what is happening.

The cable was brand new and looks clean, so I can only assume it's fine.

Yep, that's all fine. "ethtool -p" is supposed to visually identify the adaptor for an operator to see, and this makes the unmarked LED on the card blink, so the drivers are loaded fine. Interfaces are up, and taking them down and up again doesn't change anything.

Cheers, Adam.

Okay, I'll try that and see if it makes a difference. Thanks for all the info!

That does sound like the cable will be curved quite a lot though - that won't stress/break the fibre will it?

Cheers, Adam.

A couple thoughts -

*) What happens if you connect a card in loopback?

*) Have you actually "ifconfig'd" the interface? I have some vague recollection that some NICs might not try to establish link until they are ifconfig'd up under Linux.

rick jones

If you mean connecting the TX port to the RX port on the same card, then no, nothing happens. Tried it with both cards and both fibres.

Yes the card is up and assigned an IP address, and yes, this card does not establish a link at least "until the driver is loaded" according to Intel. But it's as active as Linux can make it, and still no link. Taking one of the cards down and up again has no effect.

Cheers, Adam.

Okay, well I tried using the IR/nightvision mode on a camcorder to check this, and it somewhat embarrasingly revealed that I'd connected the cables back to front. Turns out the blue and yellow connectors are already reversed, so I should've been connecting the blue connector to TX at *both* ends, and the yellow connector to RX. Blue at one end of the cable is connected to yellow at the other end. Never would've guessed that :-)

Unfortunately though, now that I've verified that there's IR light going through the fibre and coming out quite brightly, and verified that TX on each card is definitely connected to RX on the other card, it still doesn't work :-( I also discovered that if I don't secure the cables in quite so much the brightness drops off quite a bit, but even like that I still don't get a link - so it looks like it's not the signal being too strong.

Cheers, Adam.

I have to wonder if the cards are actually still good then.

rick jones

That's what I was wondering. Until about an hour ago. Just by chance I bent one of the fibres while I was plugging them in, and the link light came on, just for a split second.

It turns out that one of the fibres isn't quite at 100% - if I bend it a little just behind the plug the link lights will come on - but as soon as I let go and the cable flexes back again the link is lost.

Looking through the camera the brightness increases quite a lot when I bend the fibre, and then drops back when I let go. So it looks like not enough light is getting through one of the fibres normally.

Not sure of the best way to fix that though. Or even whether it would be considered a faulty cable, as plenty of light is getting through normally. Hmmmm...

Cheers, Adam.

Always think layer 1 first, eh? Yes, that cable is busted. Time to get another one.

Regards,

Marco.

You'll be all right. This stuff can even be wrapped around a pencil with no problem. Continuing with the stationary analogies, you might even be able to wrap it around a cartridge of a ball pen and still get away with it, but that would depend on stiffness of the plastic outer jacket much more than the fiber itself.

Just remember the golden rule of data comms Adam ... never look into a fibre optic cable with your one good eye.

BernieM

Hi, One thing not mentioned is that I seem to recall that you can loop a single interface with one piece of fiber.

If you have limited testing resources this may expand the options that you have.

Also I cannot recall /ever/ having receiver saturation with multimode. FDDI single mode yes, but nowadays I think that quite a few single mode receivers have sufficient dynamic range to deal with the transmitter even when it is quite close.

I think that most SC connectors come apart easily enough into seperate single connectors.