Slashdot Mirror
Cable Wars: Cat 6 vs Cat 7 vs. Cat 5e?
stone22 asks: "My company has decided to install a gigabit link. This will be initially used only for testing purposes, and on the longterm as a backbone for our corperate network. We allready decided to use copper, but what standard ? I've heard about problems using cat 7 cables (cross talk, bulky cables, non-standard connectors) so I could really use some hints from all you cabling experts out there."
While I can't insult you as well as the AC above, I do have to say that if you are seriously considering using this as a backbone pay the extra initial cash and lay fiber. You can go with MUCH longer runs, eliminate crosstalk, and generally find that life is much simplier. But if you have to have copper, Cat 5e UTP cable w/ RJ45s should be more than able to handle your GigE network.
Peace.
72 CD D7 52 D0 7E D8 47 44 91 D5 84 D1 59 F1 A9-This is my 128bit integer. There are many like it, but this one is mine.
Its good for hunting down background information about systems.
That is, if you're an expert at formulating queries. If you really want to reduce the number of "Ask Slashdot" questions that Google could answer better, then don't just gripe that it should have been an "Ask Google" question. Instead, teach your fellow users how to formulate an effective query.
Will I retire or break 10K?
You can really tell that a bunch of sysadmins and not highly-paid wire jockies read slashdot... A cable is a cable, indeed. LOL.
. as p
e 2. shtml
Seriously, in our lab tests, Cat 7 DID have some problems with a number of our servers- and this included non-home-grown stuff like NetApps and Suns. Cat 6 worked just fine. Check a couple of the links here for some more info.
Cat 6 actually did provide us with better benchmarks, btw.
http://www.siemon.com/white_papers/01-01-23-sff
http://www.networkmagazineindia.com/200205/kron
The most disturbing thing here is the general downward trend of respect in the Slashdot community. And why did the first post get rated a "4" when it posted info from a NIC manual saying NOTHING of value about cables or comparisons?
On the other hand, a search of Google probably would have found most of this out... but I suppose it's always nice to have a first-hand confirmation from someone who's actually looked at the question.
Well, whoop-dee-doo! Congratulations, you Read an FM. I'll bet the submitter did, too. Do you really think the posters let such questions through in order to substitute some loser's RTFM? We all know how to hunt down specs, standards, and manuals. If that was all there was to it, they could replace /. with a redirect to google. Such questions are not in to provide you with the opportunity to be a snotty little pissant, but for the discussion they stimulate. Here, there's access to the experience, wisdom, and judgement of hundreds (thousands?) of geeks, even if, so often, some snotty little AC demonstrates such limited thinking. Does anybody really think that all the answers are in the standards and manuals? On this topic, for instance, maybe somebody will be able to make the case that there is no reason for catVI or catVII, V works just fine, per the spec. Somebody else may be able to show real benefit, maybe just in certain circumstances, to useing more expensive cable. It is to be his backbone - maybe he should be looking at fibre instead. Whatever.
While I agree that if the submitter didn't even look at such docs as you provided in your crapflood, he IS an idiot, that has no bearing on our responses to the question, and responses to the responses. So, go get a cup of coffee, read something good, and calm down. Review the thread in a couple of days. There'll probably be something in it you didn't already know. Really. It's possible.
This is one of the most worthless article I've ever seen.
No, not because of the premise, I was actually curious what the answer was myself because I want to lay an inexpensive 1Gb network in my house at some point ("because I can").
It is worthless because of all the people who ridiculed the poster with the various RTFM and "look it up on Google" responses. Most other responses were to use something (fiber) that the poster seems to have obviously ruled out (maybe cost, maybe convenience, doesn't matter), so those don't help much either but at least they were trying.
He wanted to ask the opinion of his peers, not rely on what a manufacturer said or possibly out of date material. Most of the responses trashed him.
Why do people bother to take the time to produce non-responsive or patronizing answers? If you don't like the question, or you think he should spend his time elsewhere to get the answer, then simply don't post.
A lack of response is a much better way to get someone to go away than to waste your time writing and the time of everyone who pops in to see what the consensus was.
It is more productive to voice thoughtful opinions (reply) than to judge (moderate) others.
Try here
First, you don't give enough information in the article for a really good answer. You say that you want to lay a backbone cable but, you don't give details like the distance it has to traverse or the environment it will be in. You also fail to say how many nodes might be directly attached to this backbone. These are important details in the decision process. That said, read on....
I would have to agree with some of the other posts, you should use fibre. The other posts don't say why you should though so, here's why.
First, fibre is not going to be a lot more expensive than good copper. Sure, it is a little more but, the extra cost is worth the benefits. Benefits include increased distance, no interference at all and, most importantly, room to grow.
In laying a fibre cable, even a small one, you don't have a single data path, as you do with Cat 5-7. Even small fible cables usually bundle three or more pairs in the cable. That means that you can VERY easily double or triple your bandwidth in the future by lighting a second pair. Or perhaps you need a completely separate data path for some other service like maybe you want to interconnect a couple of legacy PBXs, or a video conferencing system, or a security system, who knows what.
The next thing is that fibre gives you even more room for growth. Sure GigE is great but, will it meet your needs in the long term. Already 10GigE is a reality and 40GigE is well on it's way. These can easily be implemented in the future, if you have fibre. I doubt however that Cat 5-7 will ever run 10Gig and definitely not 40Gig.
There is also a technology called Dense Wave Division Multiplexing (DWDM) that uses multiple lambdas or wavelengths on the same fibre pair. Using this technology it is possible to have 64 data paths on a single pair of fibre, that's 128Gbps aggregate bandwidth!!!!! That's all over fibre. With Cat 5-7 though, you will never have more than 1Gbps and only one datapath.
Fibre is definitely the way to go for a backbone solution. I hope this helps.
I design and test the physical side of ethernet for a living, so I think I state with all honesty and some authority that using anything more than cat-5 cable is wasting money. Kinda like buying 94 octane when your car is designed for and runs fine on 87. It has more capacity than you will ever use. Ok, true cat-5 cable may be hard to find, so when I'm saying cat-5, I'm also mean cat5e. I agree with the other posters that fiber is the way to go for a backbone, but copper is quite a bit cheaper, and pretty reasonably priced. According to the Ethernet spec, fiber is actually held to a higher standard in terms of bit-error rate (fiber is 10^-12, where copper gigabit is 10^-10- I'm pretty sure, I have them in spreadsheets to check against, but anyway, fiber is better) On the other hand, on the interfaces I've tested, the BER on maximum loss (copper) cable is usually far better than 10^-10 so that isn't much to worry about.
Be careful if you are thinking about installing fiber for possible use for 10GbE- there are a bunch of standards, and most of them seem to be incompatible with most current types of fiber, such as requirinig very small diameter single-mode fiber. At the moment the 10GbE world appears to be dominated by the long haul guys, not the LAN manufacturers, so cheap connectorization/fiber is not necessarily high on their priority list.
Remember to keep the length under 100m (as it says in the spec) and don't go through a lot of patch panels (since each connector adds loss). If you are going for maximum length, be very careful how you cut and crimp the cable- the more you can maintain the twist in the wire the better, and the more matched each wire in the pair is, the better.
Interesting fact: Since the loss of cat-5 cable is not well defined per unit length, The test cables (for 100Base-TX) are not specified in terms of length, they are specified in terms of loss. The maximum length cable that you test to is not a 100m cable, it is a 10dB loss at 16MHz cable. With good quality (cat5e) cable, that works out to around 135m.
You'd really be better off going with fiber for gigabit. Yes, Cat5e (and regular 5 if terminated correctly) will run gigabit, but there are some idiosyncracies that you might run into. Your NIC manuals should specify the requirements, I can't remember them off the top of my head. Also, we've found that different vendors support gigabit in slightly different ways on copper. We can't get SMC 1000-BaseT and Cisco 1000-BaseT to talk to each other. We ended up swapping out all our 1000-BaseT links to fiber and haven't had any troubles since.
So, go fiber if at all possible. If you absolutely have to use copper, use 5E cable from a reputable vendor (Belden, Berk-Tek, Mohawk, etc.) and use GOOD jacks and patch cords. I prefer Panduit and spec it for all our jobs, but others also make good stuff. Don't scimp on the patch cords either, these cheapies that you find many times don't test real good. Go for 5E rated patch cords with the short plugs and gold contacts.
Jason
"FORMAT C:" - Kills bugs dead!
Many of these are on the second or third revision of the card. I have found the "Tigon 3", Broadcom 57XX (5701) (tg3 and bcm5700, supported in FreeBSD, Linux, and others) 3COM 3C996 (SX and T) to be a very good card, the best of the bunch, as it has advanced packet coalescence, checksum offloading, and has the least number of interrupts with even insane amounts of malformed/attack ingress traffic. The medium seems to make little difference in the short haul.
I have also seen single mode cards for PCI, and I have also been working with single mode POS OC3, OC12 and OC48 cards for PCI.
POS OC3/OC12 for PCI here , Lucent OC12 and OC 48 cards here, just to name a few.
So, with OC48 being 2.5Gbit/sec, I think PCI/PCI-32/33/PCI-64/66 and PCI-X 133 have all seen their fair share of gigabit speed. Most of the cards listen above work rather nicely.
Now, one should use fiber wherever possible, especially for longer hauls. I have OC3 long haul cards for a 7507 at work that are rated to go 80Km in single mode. Multimode fiber transceivers and go up to 500 meters. Consumer grade fiber cards can go up to 10,000 meters as indicated by this 3COM article.
The point? Backbones are best done in fiber. Most switches support fiber, often they have removable transceivers or cards that let you pick single or multimode. I think that its easier to guarantee throughput with fiber as well, as RF and other interference doesn't play a role, and more often than not you aren't even coming close to the limits of the fiber in terms of distance.
Copper GigE is a good cheap fast short haul way to get servers hooked up to your switches. I have never had any problems using regular CAT 5e, and CAT 6 cabling demands a premium and isn't clear what the benefit is in terms of throughput. As far as better "CATS", I don't think they Spec for CAT 7 or any others has even been drafted, so its mainly marketing drivel at this point.
You will be surprised to see that these cards can all feed PCs far more information than they can take, and you will often see disks trying to keep up, and in certain interrupt driven kernels, if you put the adapter in promiscuous mode (we do this to analyze traffic) you can create kernel live lock because the driver desperately needs to poll the input to prevent userland CPU deprivation.
Again, not using Fiber for backbone is not a good idea. The cost differential is not as bad as it used to be, and you can by most any length of pre terminated SC and LX fiber. In fact, the interesting thing about fiber is the length of the cable barely affects cost. A 10 meter cable and a 100m cable are usually very close in price. It's the endpoints that cost the cash.
Legalize the constitution. Think for yourself question authority.