The internet IS a series of tubes. Kinda: A Reg 101 guide to cabling There are so many types of cables and connectors it can be confusing when you are building a data centre. I’ve taken a look at the pros and cons of each so you can decide which is the best option for you. The 10GbE standards were first published in 2002 by the IEEE and since then have been adopted gradually for use in data …
This article firmly places itself outside both categories; it's too little tongue-in-cheek to be a spoof and the subject is quite simple for most readers here. Many claims in the article are patently absurd, it's not just you (or anyone else) being dumb. And the article could use a little copy-editing too.
Most of the paragraphs contain errors or misleading information. I don't really want to spend the energy pointing them out unless someone really wants it. (And it's a little too much for just a "Send corrections" email...)
It started out with this:
> Unlike the other Ethernet standards, 10GbE provides the only
> full-duplex, point-to-point links usable for connecting
> network switches.
There have been "full-duplex point-to-point links usable for connecting network switches" since the very first actual full-duplex switches. The first I remember was 100 Mbps, but I'm pretty sure people were doing full duplex links between switches before that.
When using pluggables (e.g. SFP) the type of cabling you can use depends on the type of pluggable, not the switch. The pluggable acts as a modularization boundary. There are of course things that would need a converter (using SC connectors with a SFP-only switch) but that's got nothing to do with the switch. The only other common plug is the 8P8C, and here you have to choose between Cat6, Cat6a and Cat7. Or whatever comes next. But still the switch doesn't care.
10GBASE-T also runs on plain Cat6, though only up to 55m. The article makes it sound like you need Cat6a or better.
MMF is more expensive than SMF, not the other way around as the article states. SM transceivers are more expensive to produce though, but they have steadily been getting cheaper over the years.
10GBASE-SR doesn't have "the smallest form-factor". It's not a physical specification. SFP+ is though, and most of the 10GBASE standards are implemented by SFP+ transceivers. (Not 10GBASE-T because of power requirements and similarly not some of the more exotic long range optics.)
In the table it says CX4 supports up to 20G. It doesn't for Ethernet, where it's only 4 x 3.125 Gbps signaling and an actual capacity of 10 Gbps. But then the table of course lists a lot of things that aren't Ethernet at all.
Recommending using MMF for new installs is probably somewhat a matter of taste, but since SM transceiver costs are way down and since SMF is much more future proof I would certainly recommend using that for new 10G installs over anything MMF.
While the article mentions 10GBASE-T many times is does not list the important disadvantage of power consumption. It has improved over the years but is still around 5-6W per port, compared to about 1W per port for SM LH/LX ("10km") transceivers.
Thanks for the corrections. I've been through these with the writer and a second source, and made tweaks. Not every change you've suggested has been made, mainly because they're outside the scope of this intro to cabling, but I hope all the factual problems have been fixed up. Let me know if not.
It's worth pointing out the problems caused by a misplaced 'the' changing the meaning of a sentence:
"Unlike the other Ethernet standards, 10GbE provides the only full-duplex, point-to-point links for connecting network switches"
v
"Unlike the other Ethernet standards, 10GbE provides only full-duplex, point-to-point links for connecting network switches"
C.
"The difference between SMF and MMF is that with SMF, the light follows a single path through the fibre, while MMF uses multiple paths."
This bit is wrong/misleading. The diference is modes supported, not paths. OMx then defines the modal bandwidth of the multimode fibre which affects the distance, and also future proofing an installation. So OM1 & OM2 don't support 40Gbps or 100Gbps.
"Both SMF and MMF come in two wavelengths, 850nm and 1300nm, with the latter supporting longer distances."
SMF supports many more wavelengths, hence why it's used for DWDM. 850nm isn't generally used AFAIK, but that's more to do with SM v MM. Pretty much all common SFP's @850nm are intended for MM fibre. If you connect those to SM fibre there's a mismatch between expected core size and although it may work, it's generally a bad idea.
"The most widely used connector is the SFP+"
An SFP is a transciever, not a connector. Connectors would be FC, LC, SC, E2000. Important because they're the bits on the end of the fibres and connectors on SFPs or other kit. LC and E2000 probably most useful on an ODF/patch panel given connector size & density, and for E2000, built in dust caps. Patch cables can have different connectors each end.
"If you need to get 10GbE in longer distances, but not quite in the miles range, using MMF cables such as OM3 or OM4 with SFP+ connectors would be a good option."
I'd suggest that's a bad option. Key thing is to come up with a plan and stick to it. So OM4 offers the best speed/distance combo for local connections. If that's within a smallish datacentre, stick with that so you can do 100G if/when you need it over a typical room-sized distance. Get color coded patch cables made up, stick samples on a board somewhere prominent and set your BOFH on anyone who doesn't follow it. This will make future maintenance easier.
For longer distances, ie inter-room, inter-floor or inter-building, use SM fibre. Or make sure you run plenty of spare fibres. OM4 will support 1x100Gbps. If you need more, or want to segment then the ability to add a WDM or DWDM mux gives you more flexibility in the future, and most of that kit is designed to work over SMF.
"Thanks for the corrections."
Is it just me being under a 20+ year misapprehension? I'm fairly sure that twin-ax is a co-ax cable with two cores and a braided screen, while twisted pair is the cat5/6/7 stuff, multi-core unscreened. I think my brain switched off after reading that bit assuming the rest of the article was an April fool.
This post has been deleted by its author
"is your internet connection ... as the guy who's in charge of our "broadband" (the term is laughable) 25mb/s ought to be enough for everyone."
10 Gbps may not be needed in the home, but surely business users can get more than 25Mbps?
I'm seeing 1 Gbps small business plans go for $250/m for 1000 GB and a static IP.
I'm not sure if this is a troll, or a genuine question but I'll reply nonetheless.
This article is mostly covering datacentre --> datacentre requirements and inter-rack/inter-floor cabling requirements.
To describe a very basic scenario; If you have several racks full of servers and they're linked to a SAN with high throughput, for instance, you're going to want more than Gbit connectivity between your rack of servers and your SAN, otherwise that's going to bottleneck pretty damn quick. 10GbE is becoming commonplace in even some of the SMB environments I manage let alone the bigger ones and that's only going to steadily increase as technology progresses, certainly with SSD's becoming more affordable in small-scale storage arrays - Gbit simply doesn't cut it any more!
Currently trying to untangle 10Gb/s in rack bundled "appliance" connections that are currently using 1Gb/s LACP'ed copper uplinks to Cisco EOR switches that will be migrated to Nexus7k 10Gb ports. But they put the damned Nexus in another @#$@#% room. Can't get a straight answer out of anyone on how the hell we're gonna connect the damned things.
Can't get a straight answer out of anyone on how the hell we're gonna connect the damned things.
A Nexus 7000 is a big ol' beastie to buy without someone planning that! Depends on the appliance and the Nexus model but there's no doubt ways and means with the appropriate use of someone's cheque book.
Is the appliance being migrated to 10GbE or is that staying at 1GbE copper and just needs to be connected to the Nexus somehow?
effectively the 7k is 16 months late. It was supposed to be in play when the "appliances" landed. We'll migrate from 4*1Gb/s uplinks to 2*10Gb/s uplinks through the 7k when it goes live.
And its being planned, but the planners aren't talking to the doers.
(Hey, Reg, We need a 'violently removing follicular appendages' icon)
...a normal SFP (not SFP+) transceiver that tops out at 4Gb - bearing in mind that 4Gb implies fibre channel not ethernet anyway - can magically be used for 10GbE? Mmmhmm.
Although nothing new to many of us I'm sure the concept and effort of this article is appreciated overall nonetheless. The execution of said article? Not so much. That said some excellent points from Peter and good to see El Reg taking them on.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
