Ethernet. Burn?

[PerranOak]

So, I now have fast (for us) Internet. I've clocked the d/l speed on the iPad at 450Mb/s, tops.

On my desktop, the Wireless Adapter is a Qualcomm QCA9565 802.11b/g/n that should connect at 144/72 Mb/s though I understand that I need to halve these for comparison purposes. However, I routine get only 50Mb/s download on it even though it's just ten feet and a thin plasterboard wall away.

Anyway, I though it would be fun to run and Ethernet cable (via neat wall sockets) between desktop and router. When looking at the Ethernet controller, Windows 10 reports it as a Realtek PCIe GBE Family Controller - no other detail. Looking this up it seems that it can connect at 1000Mb/s which would be good even if I have to halve that?

However, there is some question on the web about whether they really can run this fast. I don't want to cut holes in the walls and get no better speed!

How can I verify the likely connection speed before committing?

Cheers.

[wraper]

Wired connection provides almost all of the rated speed and in both ways simultaneously.

[PerranOak]

Great, cheers.

BTW is the correct socket this one (from Amazon): “VCE CAT 6 Universal Network Socket rj45 tested to the EIA/TIA standard”?

[AndyC_772]

Plug in a cable and try it!

Provided the cable isn't too long, the link should come up at 1 Gbps. It doesn't get slower with longer cables, unless you're right at the limit of the maximum cable length your equipment can support (should be ~100m in theory). Just grab yourself a cable that's long enough to run from your PC to your cable modem, and see if you notice a worthwhile difference. If you do, then it's worth installing the wiring, and if not, then it isn't.

I've just been through a similar exercise myself, and now have an Ethernet switch in my loft which provides a wired 1 Gbps connection between the office (where the cable modem is situated), my lab, and my lounge - the 3 places where I have bandwidth hungry equipment. The switch is powered via PoE, so no need for a mains socket in the attic.

[David Hess]

Wired connection provides almost all of the rated speed and in both ways simultaneously.

I regularly get 100 MB/s or close each way simultaneously on my 1 Gb/s wired Ethernet network with a variety of adapters.

[AndyC_772]

There must be a bottleneck or other issue, then. A 1Gb/s link really, really is always 1Gb/s on the wire - it's not physically possible for it to be anything else. Gigabit Ethernet isn't "up to" a gigabit, it's a gigabit or no connection at all.

The first thing I'd do is check that the link layer really has come up at 1 Gbit. If you're using a cable that's on the limit in terms of signal integrity (very long, poor quality, damaged or mis-wired), the PHY in your PC and/or connected device might fall back to 100BaseTx.

There's no guarantee that any particular device is able to keep up with the speed of the wire. Are you connected to a switch that has other devices also plugged in? Could there be a significant amount of other traffic that the switch's CPU has to cope with? Not all switches are "non-blocking", ie. traffic between one pair of ports doesn't affect traffic between others. Some switches, especially older ones, have an aggregate throughput limit which is less than (number of ports) * (max link speed per port).

If that's not it, then maybe the device you're ultimately talking to just isn't that fast, and the Ethernet link speed isn't the issue.

It could also be a protocol issue; a badly configured MTU somewhere can cause a great deal of packet fragmentation, retransmission and loss, all of which can cripple the useful throughput of the interface. A session with Wireshark could be time very well spent to see if this is happening on your network.

[fordem]

There must be a bottleneck or other issue, then. A 1Gb/s link really, really is always 1Gb/s on the wire - it's not physically possible for it to be anything else. Gigabit Ethernet isn't "up to" a gigabit, it's a gigabit or no connection at all.

If you're responding to the post directly above - please note the change in units - 100MB/s is pretty close to 1Gb/s.

Having pointed that out - having a 1Gb/s link connected at 1Gb speed does not guarantee 1Gb/s throughput - you need to look at the source & destination of the transfer - I've watched transfers to servers with RAID5 arrays clock 800 Mb/s on small transfers (less than the RAID card cache size) and drop to 200~300 on larger transfers (after the cache fills up), and one of the consumer grade NAS I owned a few years back had a 1Gb/sec interface and could not transfer data faster than around 180 Mb/s because the disk interface was simply too slow.

Last - in my experience - a 1Gb/s NIC does not guarantee Gb/s speed, even in cases where the source & destination are not the bottleneck - I tend to avoid Realtek products because they just don't deliver - 1 Gb/s add-in NICs with PCI interfaces could not handle more than 120 Mb/s - swapping them out for a 3Com card allowed roughly double that, I also owned a "business class" desktop built by one of the top three manufacturers whith 1Gb/s LOMB that delivered less than 100 Mb/s, and if you've every used a USB-to-Gb/s adapter, you'll find they are limited by the available bandwidth on the USB side.

[David Hess]

There must be a bottleneck or other issue, then. A 1Gb/s link really, really is always 1Gb/s on the wire - it's not physically possible for it to be anything else. Gigabit Ethernet isn't "up to" a gigabit, it's a gigabit or no connection at all.

Packet loss can cause variable data rates.  Connections between my old Netgear GS108T switch and my newer TP-Link TL-SG108E switches had packet loss problems which I could only resolve by replacing the Netgear.

I have not tested it in detail myself but with Windows 10, the continuous scanning of file copies by the security system has a major effect on the maximum throughput.

and one of the consumer grade NAS I owned a few years back had a 1Gb/sec interface and could not transfer data faster than around 180 Mb/s because the disk interface was simply too slow.

My old Netgear ReadyNAS usually gets about 30 MB/s but sometimes holds 100 MB/s for 10 or 15 seconds so I am not sure why.

[wraper]

There must be a bottleneck or other issue, then. A 1Gb/s link really, really is always 1Gb/s on the wire - it's not physically possible for it to be anything else. Gigabit Ethernet isn't "up to" a gigabit, it's a gigabit or no connection at all.

Packet loss can cause variable data rates.  Connections between my old Netgear GS108T switch and my newer TP-Link TL-SG108E switches had packet loss problems which I could only resolve by replacing the Netgear.

Packet loss in ethernet means you have a faulty connection. Not necessarily it was Netgear issue, it might be a too long or bad cable. And another device was more tolerant to it.

[David Hess]

Packet loss can cause variable data rates.  Connections between my old Netgear GS108T switch and my newer TP-Link TL-SG108E switches had packet loss problems which I could only resolve by replacing the Netgear.

Packet loss in ethernet means you have a faulty connection. Not necessarily it was Netgear issue, it might be a too long or bad cable. And another device was more tolerant to it.

All tests showed a problem between the Netgear and TP-Link switches no matter what cable was used, and no problem with the cable unless the Netgear switch was using it.  It may very well be that the Netgear and TP-Link combination was less tolerant of cable issues, but no cable issues were found with any other combination and the best solution was to replace the Netgear switch.

Beyond observably poor throughput, these switches all have management functions which allowed tracking bad packets making diagnostics unambiguous.

[PerranOak]

Is CAT6 the one to get?

[David Hess]

Is CAT6 the one to get?

For 1000BASE-T (Gigabit Ethernet), CAT5e is sufficient but given the small if any increase in price now, I do not think there is any reason not to use CAT6 or CAT6A.

[madires]

That question isn't easy to answer. It depends on what your are planning for the future. CAT5e is fine for 1000Base-T and 2.5GBase-T. 5GBase-T works too with shorter cable runs (<70m?), also 10GBase-T (<50m). CAT6 is a safe bet for up to 5GBase-T, while also limited to 50m for 10GBase-T. For running 10GigE at full segment length you'd need CAT6A. Be aware the higher the CAT rating is, the more effort it takes to install it.

My personal opinion is that TP wire is fine for 1GigE and the new 2.5/5G "power ups". Short cable runs at home might be OK for 10GigE, but for anything beyond that I'd use fiber (multimode OM3 or 4).

[Ian.M]

Beware of CCA and CCS cable (copper clad aluminum and steel).  Neither are fit for purpose as network wiring.  The easiest way of spotting them is with a magnet for CCS and by flame test for CCA.

To flame test solid core ethernet cable, strip a wire totally and hold the core with pliers with 1" protruding past the flame.

You need pure copper CAT5e or better cable. Make sure its described as copper and if it isn't return it!  Suspiciously cheap cable is rarely genuine.

[David Hess]

My personal opinion is that TP wire is fine for 1GigE and the new 2.5/5G "power ups". Short cable runs at home might be OK for10GigE, but for anything beyond that I'd use fiber (multimode OM3 or 4).

I would actually avoid 10GBASE-T (10G Ethernet over CAT6 and CAT6A cable) because of the high cost of the hardware compared to SFP+ interfaces which can use relatively less expensive copper cables for shorter distances and fiber for longer ones.  Maybe 10GBASE-T will eventually come down in price but right now it is the more expensive solution.

[PerranOak]

Sorry, I don’t understand half the words used above. 

The cable between sockets will only be about 6m long. However, if I had to use a made-up cable to test it it would be more like 20m - sounds expensive.

[Ian.M]

The cable between sockets will only be about 6m long. However, if I had to use a made-up cable to test it it would be more like 20m - sounds expensive.

Unless you know someone in the network installation business who can do you a deal on the remnant of a reel, or better yet can do the whole job at 'mate's rates', your first problem will be finding a reputable supplier for Cat6 cable by the metre, otherwise you'll be stuck buying a 50m reel for a price not much less than a 100m reel.

[Syntax Error]

Perran, I think as David Hess was eluding to, over very short distances < 10m, 1000baseT works just as well with Cat5 as Cat6 rated connectors/cable. In practical terms for a domestic network, you will notice zero difference. Afterall, the internet protocol is designed to deal with dropped packets. But in 2020, cat6 parts only cost pennies more than cat5. Choose cat6 because it's future proof.

If you want Cat6 cable, connectors, etc, pop over to ebay. Plenty of UK sellers selling network leads of all types and lengths. And check out cool *flat* cat6 leads, which are ideal for running under carpets and behind TVs.

If you don't already have one, buy a RJ45/RJ11 network cable continuity tester - with the green leds. It will save you hours of stress for years to come. Around £10-15.

[madires]

I would actually avoid 10GBASE-T (10G Ethernet over CAT6 and CAT6A cable) because of the high cost of the hardware compared to SFP+ interfaces which can use relatively less expensive copper cables for shorter distances and fiber for longer ones.  Maybe 10GBASE-T will eventually come down in price but right now it is the more expensive solution.

I fully agree. However, at home some people prefer to keep using their old wiring and spend more on active components than thinking about a new wiring.

[madires]

Perran, I think as David Hess was eluding to, over very short distances < 10m, 1000baseT works just as well with Cat5 as Cat6 rated connectors/cable. In practical terms for a domestic network, you will notice zero difference. Afterall, the internet protocol is designed to deal with dropped packets.

CAT5e! That little 'e' is quite important. And you can run 1000BASE-T on up to 100m CAT5e. Of course, you can also use anything better, like CAT6, but you'll pay more without gaining much for home use. BTW, if an ethernet link is broken TCP/IP can't do anything about that.

[Monkeh]

Perran, I think as David Hess was eluding to, over very short distances < 10m, 1000baseT works just as well with Cat5 as Cat6 rated connectors/cable. In practical terms for a domestic network, you will notice zero difference. Afterall, the internet protocol is designed to deal with dropped packets.

CAT5e! That little 'e' is quite important. And you can run 1000BASE-T on up to 100m CAT5e. Of course, you can also use anything better, like CAT6, but you'll pay more without gaining much for home use. BTW, if an ethernet link is broken TCP/IP can't do anything about that.

You don't need the 'e' for gigabit at 100m - it was originally specified for cat5.

[madires]

From a specification standpoint you're right and many CAT5 cables even meet the CAT5e specifications. From a practical standpoint things look a little bit different. Back then when GigE was standard in data centers we saw issues with long runs of some Cat5 cables and those cables met CAT5 specs (tested). CAT5e performed more solid. Also, CAT5 is obsolete for many years.

[Monkeh]

From a practical standpoint things look a little bit different. Back then when GigE was standard in data centers we saw issues with long runs of some Cat5 cables

Crosstalk issues in cable bundles have approximately no relevance to normal home usage.

[madires]

There's also crosstalk between the four pairs in one cable.

[Monkeh]

There's also crosstalk between the four pairs in one cable.

Well, this is true. Again, house, short runs in almost all cases, and also, as you pointed out, where does one find cat5 these days?