14

My work from home office, in a difficult-to-plumb part of my house, has an existing CAT3 running to it from the main part of the house. I wanted to use a wired connection because the Wifi flakes out occasionally, but it would cost a lot to properly run a CAT6 cable. While we're considering that, I opted to just use the CAT3 cable because I'd rather have a slow but solid connection instead of a fast but sketchy Wifi connection.

Despite CAT3's low transfer rate specification, I find I am able to get the same speeds as using the CAT6 cable directly on the router.

enter image description here

I use this connection all day at work and it is always this fast (I do a lot of data transfer for my job), so I know the speed test is not just a fluke. Other posts specifically say forget Gigabit networking on CAT 3, which is in line with other sources I've found.

Can someone help me understand how a CAT3 cable is able to consistently transfer data this fast? The answer will help me understand whether I should consider the upgrade to CAT6.

user1717828
  • 263
  • 2
  • 6
  • 37
    You got lucky. The CAT3 cable (and its installation) is exceeding its specifications. How long is the run? – Dave Tweed Aug 02 '21 at 13:04
  • 2
    @DaveTweed I can't be sure of the length because it runs through inaccessible parts of the house, but it's a minimum of 100 feet. – user1717828 Aug 02 '21 at 13:18
  • 2
    I gather it is UTP cable, I would expect 25 dB attenuation/100m at 100MHz or 8 to 10 dB loss at 100 MHz and assume it is HDX. I get the same 1Gbps using cheap cable in the basement – Tony Stewart EE75 Aug 02 '21 at 13:26
  • Got a link on cable+ MOBO specs? Twists /ft? OD ratio ins/conductor determines 100 ohms but R/100’ affects group delay distortion which if card does equalization well can improve performance – Tony Stewart EE75 Aug 02 '21 at 15:11
  • If you look at TIA-EIA-568, which is the industry specification for network cabling, you will see that Cat 3 cable's performance is not even specified above 16 MHz. Loss (attenuation), FEXT, NEXT, etc numbers all stop at 16 MHz. Return loss is not specified at all. – SteveSh Aug 02 '21 at 17:46
  • So I agree with David Tweed - you just got lucky! Sort of like using a 3.3V part at 4 V. – SteveSh Aug 02 '21 at 18:02
  • I've found that Gigabit Ethernet is pretty robust. I've run it through a series of three connectors meant for power and still worked fine for distances up to 50 metres. – qrk Aug 02 '21 at 23:17
  • 5
    Gigabit ethernet wired PHYs are ingenious feats of engineering, and are very complex digital signal processors that perform billions of multiplies-and-additions per second IIRC. They can extract bandwith from a given cable much better than 100TX PHYs can. But, just because they "work" doesn't mean that there's much margin left :) – Kuba hasn't forgotten Monica Aug 03 '21 at 06:19
  • 2
    If the CAT3 is already run, can you tape your CAT6 cable to one end and use the existing cable as a wire pull? (if cable clamps were not used in the installation) – Dan A. Aug 03 '21 at 17:46
  • 3
    I think it has transitioned. Don't ask questions, don't touch it, and only address it as "CAT 3/4/5/6+" and it may continue to function. – Peter - Reinstate Monica Aug 04 '21 at 14:05
  • 2
    @DanA. That usually doesn't work. The original cable is often stapled, or punched through a hole that isn't big enough for the 2 cables bundled. Also very likely, the twists and turns will bind too much to place meaningful pulling force on the wire. – Harper - Reinstate Monica Aug 04 '21 at 18:47

5 Answers5

36

100 feet is not extremely long; so, as Dave says, you've got lucky, and your Gigabit Ethernet hardware is just exceedingly robust.

You have to realize that "Cat3" doesn't put an upper limit on the bandwidth that's usable – it's just guaranteeing that signalling bandwidths of up to 16 MHz work. Now, it seems pretty unlikely that classical Cat3 "telephony" wiring does Gigabit ethernet¹; that starts with Gigabit ethernet requiring four pairs, and Cat3 rarely being installed with all four pairs connected.

So, maybe you also got lucky with respect to your installation, and it's not actually Cat3, but something "labeled down" to Cat3. Was it installed after ca 1998? I think that was an era when Cat5 cables became ubiquitous, but I guess some people still demanded Cat3 (because that was what they were using before), so some manufacturers simply printed Cat3 on Cat5-capable cables (and sold them at a markup).

¹ hard to quantify, but we should be seeing serious attenuation here for the higher frequencies, requiring equalization from both ends of the link that far exceeds what the standard says devices need to be capable of.

Marcus Müller
  • 88,280
  • 5
  • 131
  • 237
  • People demanded Cat3 over Cat5? Was there a side-effect or even a suspected side-effect of Cat5 that caused this? I've seen people avoid Cat6 because of the shielding issues when combining Cat6 with Cat5, was it something like that? – Mast Aug 03 '21 at 17:01
  • 5
    No, it's more like people have decades of rules to follow that said "Cat 3". – Marcus Müller Aug 03 '21 at 19:25
  • 9
    @mast As Marcus says some (years old) building standard said CAT3 so you had to install CAT3 even though it was already way behind the times. In the mid 90's we installed cable that came on 1000 feet rolls. It had (as usual) the CAT label stamped on the cable every 3 feet or so. Our supplier helpfully alternated CAT3 and CAT5E markings along the cable. Depending on how clueless the standards-inspector or the customer was we showed them a piece of sample cable with the CAT3 mark or another piece with the CAT5E mark. – Tonny Aug 03 '21 at 22:58
  • 3
    @Tonny To be fair, a CAT5E cable _also_ up to a CAT3 standard, so... :) – Luaan Aug 04 '21 at 07:18
  • 5
    @Luaan You know that. I know that. But try to explain that to a auditor that doesn’t know anything about tech and is just interested in checking of all the checkmarks on his list. “List says CAT3 so you better show me the CAT3 label” – Tonny Aug 04 '21 at 07:41
  • 2
    also note that one of the oldest tricks in the book is "look, foreman, we're putting these *much better stuff* in here, you're getting them for the price of the regular stuff you specified", and then putting in whatever was left on stock from your last job with zero consideration for appropriateness, and afterwards of course still invoicing as if you had to buy the proper material. The job of auditors and foremen is making sure things are non-corrupt and up to spec. In very rare cases they are *allowed* to deviate from specification! – Marcus Müller Aug 04 '21 at 08:15
14

The Ethernet speed negotiation process takes into account the capabilities of the devices at each end of the connection, but not the cable between them. Essentially, this means that if you connect two gigabit-capable Ethernet devices with a cable that's not good enough for that, they're going to go ahead and try to talk to each other at gigabit speeds... and get a lot of transmission errors as a result.

So the surprise here isn't that the devices are sending at gigabit speeds, it's that you aren't getting a lot of packet errors (and retransmissions). But if possible, I'd take a look at the interface statistics on your computer (how to do this depends on the OS), and see just how many garbled packets you're getting.

Edit: from the comments below, at least some Ethernet interfaces will negotiate down to a lower speed if they detect that the cable isn't up to the original speed (I was obviously not familiar with this feature). So either your interfaces don't do this, or they do & didn't detect enough of a problem to drop down. I'd still check for errors if you can.

Gordon Davisson
  • 249
  • 1
  • 5
  • 4
    this is nonsense. The negotian process downrates the connection speed to get the errors to an acceptable level – camelccc Aug 03 '21 at 09:32
  • 2
    @camelccc: AFAIK, downshifting isn't officially part of the standard negotiation -- although it's indeed so common that I *thought* it was standard, it still turns out to be a vendor-specific feature (e.g. Intel seems to call it "SmartSpeed" and it's implemented at driver level). – user1686 Aug 03 '21 at 10:17
  • 1
    _[That being said, the Gigabit Ethernet spec does seem to take some cable properties into account, e.g. it claims that it's able to handle arbitrary pair mixups and accidentally reversed polarity, which I think is interesting.]_ – user1686 Aug 03 '21 at 10:22
  • 4
    @camelccc, I find that hard to believe, since I've seen gigabit-capable devices negotiate a gigabit link over a two-pair cable. With the obvious result of managing to pass no data at all. – ilkkachu Aug 03 '21 at 17:46
  • 1
    I have seen exactly the problem described in this answer -- two gigabit transceivers happily negotiated a gigabit link over a marginal cable, resulting in huge numbers of errors and an unusable link. So I'm inclined to believe it. – Glenn Willen Aug 04 '21 at 01:40
  • 1
    @ilkkachu If you are going to ask for a list of silly things devices do, it is endless. sometimes you get 2 gigabit devices that talk to any other device at gigabit speed, but not 2 each other, instead downshifting to 100M. 95% of the time though, devices negotiate the maximum usable speed, its because of the the 5% the time they don't that people turn off autonegotiation. Generally with gigabit you can go way outside the standards with no iff effect. Even anything over 10 feet on a stranded cable is technically outside. 10G Tbase on the other hand I have always found finicky, – camelccc Aug 05 '21 at 12:15
  • @camelccc, I didn't ask for a list of all the silly things devices do. I'm just saying, by way of an anecdote, that expecting standard 1000Base-T devices to actually test the cable qualities (other than identifying the pairs) is wishful thinking at best, and claiming the answer here is "nonsense" seems overly bold. Maybe some devices do it, but the understanding I've had is what user1686 said there: that it's not a standard feature. But without someone actually quoting the relevant standards, I'm not sure we're going to get anywhere. – ilkkachu Aug 05 '21 at 13:17
  • I took and look before and after downloading another 1 GB and the there was no change in packet errors (using MacOS). This doesn't surprise me because I regularly download large datasets for work and local hashes match cloud matches and the downloads don't take longer than expected by the speed test rates. – user1717828 Aug 05 '21 at 21:29
9

Check the error counters on your attached equipment - managed switches generally keep a track of errors and collisions. It may be that this link is suffering errors, but at gigabit/s, it's still retransmitting the Ethernet frames (layer 2) fast enough to be unnoticeable.

Another option is that the cable wasn't Cat 3 originally. Electricians have been known to use what they have available rather than waste time getting a lower-specification able. If you can expose a length of it and read the markings off the outside, that could be interesting. I remember a site where the cable was Cat 5 in the floor, but only Cat 3 jacks had been used for financial reasons.

If the cable is partially run inside metal pipe or some other conductive material, that may be adding an accidental element of shielding. Even strapping the twisted pair cable to the side of a metal pipe or plane can provide some protection from external interference.

Also, your test process in the screenshot might be lying or misleading. Buffering and caching can upset tests, and ISPs can benefit by providing "optimistic" results. Instead, check the link speed of your network card using commands like ifconfig in most OSes, or ethtool and mii-tool in Linux. Similar commands can be used to show interface error counters.

Lastly - don't look a gift horse in the mouth!
Enjoy your speed, and next time you have anything open/dug up, take the opportunity to drop in cables and/or draw wires to pay back Saint Murphy for staying away from your structured cabling!

Peter Mortensen
  • 1,676
  • 3
  • 17
  • 23
Criggie
  • 754
  • 7
  • 19
1

Cat 3 and Cat 5 have a characteristic impedance of 100Ω so it's not surprising that they could behave in the same way. The capacitance is different between cables because cat 5 has more twists than cat 3, but it's conceivable that you could get fast rates from a cat 3 cable.

You should upgrade to cat 5 or cat 6 where possible, because its a lot easier to replace the cable once than it is to replace it with something cheaper then find out you need to replace it again for a faster speed.

Voltage Spike
  • 75,799
  • 36
  • 80
  • 208
  • 7
    While true, there's a lot more to "how far you can communicate with ethernet" then just the impedance of the interfaces. – SteveSh Aug 02 '21 at 21:19
1

On a RF-noise clean setting you do get awesome Ethernet performance, thus if you're away from power cables and energy line, no close antennas or Wi-Fi it's quite clear the cable can perform quite well for itself depending the thermal conditions.

Also if you put the cable inside a metal protection you may get a little to no error ratio but the auto-negotiation may work, maybe not.

Jakob Halskov
  • 1,542
  • 12
  • 18
Sen
  • 11
  • 1