72

You often find Switches that say something like 10/100/1000Mbps.
I get that the numbers mean possible speeds, but why not just write "up to 1000Mbps" or something? Is there more meaning to it?

MaQleod
  • 13,149
  • 4
  • 40
  • 61
user2037559
  • 779
  • 1
  • 5
  • 6
  • It might help exactly one potential customer refraining from asking sales reps "does it support 10Mbps?". But usually it doesn't. Marketing. 'nuff said. – vautee Feb 13 '18 at 10:37
  • 39
    They are specific modes. I have a bad cable right now so network will either transfer full rate 125MB/s (1000mbps) OR 12.5MB/s (100mbps). It does not run just slow; it switches to a different mode. – Damon Feb 13 '18 at 10:41
  • @Damon so do you literally mean a bad cable (like a cat4/5) or a bad provider? – user2037559 Feb 13 '18 at 10:46
  • 30
    @Damon mbps = mili bits per second... Mbps is the standard (mega bits per second), don't quote in MB/s (mega bytes per second). – Attie Feb 13 '18 at 11:05
  • 6
    A device could, in theory, not support the lower speed standards and only operate at higher speeds. By stating all three they are being clear that they fall back to whatever is he highest supported standard. – Mokubai Feb 13 '18 at 11:20
  • 17
    I have a few switches that don't support 10Mbit, so not just in theory. – Simon Richter Feb 13 '18 at 11:23
  • 2
    @user2037559 He probably mean a bad cable. 1000BASE-T uses all 4 pairs of the cat5+ cable while 100BASE-TX only uses 2 pairs. So if one of the two pairs used for 1000BASE-T is broken then the cable can still be used for 100BASE-TX. – simon Feb 13 '18 at 13:20
  • 1
    The cable thing is a bit tricky: Having a bad cable does not force the devices to a lower standard! If you have for example a cable with just 2 pairs, which is fine for 100MBit, two GbE devices will still not work over it unless you configure at least one of them manually to use 100MBit... The negotiation is just based on the capabilities of the 2 devices and not considering the cable quality/type. – JCH2k Feb 13 '18 at 14:45
  • Bad connection of one of the conductors on one end of a cat 6 patch cable is what it seems to be – Damon Feb 13 '18 at 15:59
  • 4
    @JCH2k Your theory doesn't match practice. – kubanczyk Feb 13 '18 at 17:21
  • 1
    Note that fibre(fiber) ports don't generally support link-speeds other than their maximum. And perhaps half of copper SFPs/GBICs etc won't do speed autonegotiation either. The copper thing has caught me out more than twice, embarassingly. – Criggie Feb 13 '18 at 20:44
  • It also indicates that the switch can run _different_ speeds on each port. – Thorbjørn Ravn Andersen Feb 14 '18 at 08:27
  • 1
    @kubanczyk I had exactly that case a few years ago: GbE switch, GbE Card, but only 2 twisted pairs of telephone cable through the house. It didn't work until I set the mode of my GbE card from Auto-Negotiate to 100MBit Full Duplex, so it's not theory! – JCH2k Feb 14 '18 at 15:23
  • 1
    10BASE-T support is being dropped from devices, or at minimum being replaced by a mostly-compatible but potentially incompatible version (10BASE-Te). 10BT requires a higher signaling voltage that causes problems and higher power consumption in PHY design, especially with 10GBASE-T. – user71659 Feb 14 '18 at 21:49
  • @Attie: that [was decided in Meta](https://meta.stackexchange.com/questions/87251/use-m-for-million-instead-of-m) already. Lower case m is great. – Thomas Weller Feb 15 '18 at 10:19
  • It is always preferable to explicitely state what the device is capable of, without making assumptions. Up to 1000 Mbps could cover 10/100/1000 Mbps or 100/1000 Mbps or just 1000 Mbps. So up to X might be ambiguous. – Dohn Joe Feb 15 '18 at 23:25

4 Answers4

130

Not the same thing at all. Indeed, there exists no such thing as "up to".

10/100 Mbit/s uses the same cable (ignoring different specifications for the shielding, which is negligible from the switch's point of view). They use a different clock and might (I'm not 100% sure, don't nail me down on that one) use a slightly different signal coding.

GbE uses a much different cable (with two more copper wires) and most certainly uses an entirely different signal coding that pushes through more bits per clock. The clock rate is -- surprisingly to the unsuspecting user -- indeed not ten times higher than that of 100Mbit/s, only four times.
So, that's in principle a totally different, hardware incompatible thing, which only happens to have the same general marketing name "Ethernet", and the same kind of RJ45 plug.

The inventors of Ethernet were intelligent enough to add a very extensive negotiation capability, so it is possible (but not granted) that a device built for one standard also supports another.

Thus, while in practice a switch supporting GbE always (always? well, maybe...) supports 100 Mbit/s and 10 Mbit/s as well, it does not need to do that. It's a bonus, if you want to see it like that, and it is not "everything up to", but it's implementing three very specific, different, well-defined standards.

Nayuki
  • 712
  • 1
  • 6
  • 11
Damon
  • 4,512
  • 3
  • 22
  • 28
  • 22
    This is very true. Devices exist which support only 100Mbit/s and fixed duplex. you still find industrial devices which only support 10Mbit due to the simple signaling. – feitingen Feb 13 '18 at 12:40
  • 8
    Also, some older (typically late-1980s - early-1990s) deices that use 10Mbps will not work with most modern 10/100 or 10/100/1000 switches. They do not auto-negotiate. – pseudon Feb 13 '18 at 14:44
  • 4
    Nicely worded, I'd like to throw out that I have a WLC that is gigabit Ethernet, and only gigabit Ethernet. The switch I tried to use was 10/100. It took me a while to find out that the WLC only supported gigabit (both the switch and the WLC shows up/up on the interface). – Allen Howard Feb 13 '18 at 16:14
  • 4
    Don't you mean "two more copper wire pairs" or "four more copper wires"? – whatsisname Feb 13 '18 at 19:32
  • 1
    The last time I plugged a 10 into a 10/100/1000 I found out it was really a 100/1000. – Joshua Feb 13 '18 at 21:48
  • "might (I'm not 100% sure, don't nail me down on that one) use a slightly different signal coding." They use totally different signal coding. – plugwash Feb 13 '18 at 21:55
  • Pretty sure devices at the carrier level only support a single standard, cf https://www.juniper.net/uk/en/products-services/switching/ex-series/ex4600/ these require different modules to support different bandwiths. Granted this is fibre but I'm sure you could find something similar with copper – undefined Feb 14 '18 at 00:29
  • 17
    GbE doesn't use a different cable; it's rated for Cat5 or Cat5 Enhanced, which were existing cable standards used with 100Base-T. Unlike 100Base-T it actually *uses* all four conductor pairs for communication, meaning a faulty or below-spec Cat5/6 cable can work fine for 100Base-T if the fault is in one of the conductor pairs not used in that standard, but fail for 1000Base-T. But a cable without the 3rd and 4th pair never would have met the Cat5 standard. – thomasrutter Feb 14 '18 at 02:51
  • *"The clock rate is -- surprisingly to the unsuspecting user -- indeed not ten times higher than that of 100Mbit/s, only four times."* - this might be a bit misleading to the unsuspecting reader. :) The clock rate is exactly the same 125Mbps in 1GbE. The bandwidth is 10x bigger because it uses 4 wires of the existing Cat 5 cable to effectively transmit 8 bits per a single clock (hence 125 x 8 = 1000). The difference between Cat 5e and Cat 5 is just the stricter interference spec in the former. – vgru Feb 14 '18 at 10:59
  • @thomasrutter there is more than one GbE physical standard - the one we are mostly using uses more wire pairs, others use more bandwidth per wire pair. – rackandboneman Feb 14 '18 at 15:01
  • @Joshua maybe you plugged something that cannot autonegotiate and is 10 into a 10/100/1000 port set to autonegotiate? – rackandboneman Feb 14 '18 at 15:03
  • @rackandboneman: It was plugged into an unmanaged switch and how can something that's only 10 autonegotiate? – Joshua Feb 14 '18 at 16:08
  • @Joshua It can tell the switch that 10base-T is all it supports. – rackandboneman Feb 15 '18 at 10:17
  • @rackandboneman: It's only 10. I don't think auto-negotiate existed when it was made. – Joshua Feb 15 '18 at 16:07
  • 1
    @pseudon modern managed switches allow you to force a port to a specific speed/duplex so that they can work with such devices – bparker Feb 15 '18 at 19:53
39

They are entirely different standards - it's not like an ADSL connection that essentially does "the best it can" to get data to you at your hoped-for speed [the famous "up to" advertising lies claims made by ISPs until recent legislation to stop them] .

Each standard has a specific interaction. If you don't have the right cable, or the wiring is sub-optimal, the system will automatically switch down to a slower but more robust connection mode.

More than you'd ever need to know at Wikipedia - Ethernet over twisted pair

Tetsujin
  • 47,296
  • 8
  • 108
  • 135
  • 9
    As said, it's not "up to". If your cable can't support 1000Mb/s, it'll drop to 100Mb/s. It won't plod along managing 750Mb/s. If that were the case, networking would be a chaotic mess of different cables achieving whatever they can and you'd have to benchmark each step to find out what's achieving what to find bottlenecks. This system avoids that, and most 10/100/1000 switches will show a different light colour depending on which standard it's connected at. – i-CONICA Feb 13 '18 at 12:46
  • 15
    @i-CONICA: and audio cable manufacturers would be having a field-day selling gold-plates heat-treated phlebotenum-infused ethernet cables. – Peter Cordes Feb 13 '18 at 16:08
  • 9
    @PeterCordes They [sell those anyway](https://arstechnica.com/staff/2015/02/to-the-audiophile-this-10000-ethernet-cable-apparently-makes-sense/), because why let logic get in the way of trying to find someone who will pay $10,000 for an ethernet cable? Hilariously, it's even marked with an arrow for which "direction" you want the audio to go. – Zach Lipton Feb 13 '18 at 17:51
  • 5
    @ZachLipton: After posting my last comment, I remembered that they already did this for purely-digital video and audio cabling (e.g. HDMI), but I hadn't realized there were voodoo ethernet cables, too. – Peter Cordes Feb 13 '18 at 18:20
  • 1
    I remember the days when audio cabling 'voodoo' could be achieved by simply tying the correctly-shaped knot in it ;-) – Tetsujin Feb 13 '18 at 18:29
  • 2
    @ZachLipton: Oh lordy – Lightness Races in Orbit Feb 14 '18 at 02:30
  • 4
    @i-CONICA: It will kind of detect, but it's ugly. It "detects" by dropping packets as the cable unable to sustain them corrupts them. A 100 with 50% packet loss is still faster than a 10 if you're willing to play with UDP and rapid recovery. – Joshua Feb 14 '18 at 16:50
3

Such a switch may support up to 1000Mbps, but only through 3 distinct cable protocols. These three supported protocols are individually labelled to confirm it does indeed support those three (and none other!), by speed: 10Mbps, 100Mbps, 1000Mbps.

It is helpful to the customer/user that they are explicit, as it avoids compatibility issues (e.g. not supporting the 10Mbps standard anymore).

Lodewijk
  • 131
  • 2
0

Ethernet never supported other speeds than 10/100/1000Mbps in the 0-1000Mbps range, so writing "up to 1000Mbps" wouldn't be outright wrong. It would, however, leave customers in doubt whether 100Mbps and especially 10Mbps are supported, so all supported speeds are explicitly mentioned.

Dmitry Grigoryev
  • 9,151
  • 4
  • 43
  • 77