Testing limits of home Ethernet wiring

unmesh

Member
Apr 17, 2017
37
11
8
62
We started out with 100Mb connections in our home-run wired house and, over time, moved up to gigabit connections without incident. I recently ran a couple of 10Gb links in my home office using DAC cables and am now thinking I should test the in-wall wiring to see whether any of the cables can support 2.5/5 or even 10Gbps. The cables are Cat5e and terminated in 5e rated jacks in the network closet and 5e rated keystones in the rooms.

Any suggestions for how to do this?

I will need NbaseT/10BaseT endpoints, at least one of which should be portable to move to the different rooms. For the latter, are USB/Thunderbolt Ethernet dongles good enough or do I try to find a SFF or thin-client PC that can take a PCIe NIC?

Should I get a Cat6/6a rated patch cords to connect to the keystone or does the short length make this an insignificant factor? Is there a way to tell whether changing the terminations on the in-wall cable will enable the next higher link speed without actually re-terminating? Is terminating Cat5e cable into Cat6/6a plugs/jacks even a thing?

Is the autonegotiated link speed indication good enough as a test or should I plan on running something like iperf? Any particular iperf parameters? Is there a link error count for which the link does not autonegotiate down but at which prudence dictates manually provisioning a slower link speed or should one trust the autonegotation regime?

Lots of questions; thanks in advance for answers/suggestions :)
 

Airz

New Member
Jan 22, 2014
22
1
3
You can run 10Gbe over Cat5e as I'm doing it myself but my cable run is only about 10 metres so it's not an issue. I do use Cat7 patch cables at both ends but it worked with Cat5e patch cables before they arrived in the post. The best thing to do is just walk around the house with a 10Gbe capable device and test which sockets work and which don't with cables you have at the moment. If you find a socket that doesn't work you could then try buying a couple of Cat6a patch cables for each end to see if that resolves it.
 

unmesh

Member
Apr 17, 2017
37
11
8
62
Makes sense to start with what one has at hand and the first thing I'm missing is a 10Gbe (and hopefully 2.5 and 5) capable device that I can walk around with! Ideally battery powered :)
 

Samir

Well-Known Member
Jul 21, 2017
1,534
507
113
46
2.5Gb/5Gb should be easy enough to support if 1Gb is working fine--it's the 10Gb test that's going to stress it all.

What I did at my parents house (over 50 drops) was to simply have a system running iperf and take a laptop and run an iperf test from each jack. I would do this since just because something links doesn't mean it will transfer at proper speeds--I saw everything from 700Mbps to 900Mbps--but still not bad for wire that was installed in 1995. :)

To repeat this type of test with up to 5Gb will be easy as there are 5Gb usb dongles (even though they don't even hit 4Gbps), but for 10Gb, you practically need to have two devices--a laptop and then the 10Gb adapter. :eek: Plus in both of these scenarios you either need a capable switch or someone at the patch panel moving ports as you do.
 

sh1

New Member
Sep 20, 2020
4
1
3
Depending on the distance and if you have 2+ lines to the rooms, you can loop it back to the closet and run the tests there. If it works, great. If not, there's more work to do. I don't have any issues running 10Gb over Cat5e on a ~40' run.
 

unmesh

Member
Apr 17, 2017
37
11
8
62
I'm already running 1GbE and the standards say I "should" be OK at 2.5Gb and 5Gb but it would be good to test this out.

I found some 2.5GbE USB dongles for $20 each and bought a pair (and posted the link in the deals section). 5GbE dongles are expensive especially in my test only/mostly use case as are RJ45 SFP+ transceivers that I'm not sure will even work in my Mellanox CX311 SFP+ NICs and especially at NbaseT rates. And I recently learned that my old Juniper switch only does 10G/1G on its SFP+ ports so I'd end up doing a NIC to NIC test at 10Gbps.

@Samir, what were you thinking of when you suggested a 10Gb endoint on a laptop? And I will recruit a helper.

@sh1, all rooms have two cables behind the wall plate but only one is terminated in a keystone jack.
 
  • Like
Reactions: Samir

Samir

Well-Known Member
Jul 21, 2017
1,534
507
113
46
I would run the same type of iperf test simply patch to jack with the 2.5Gbps and 5Gbps. This will require two systems, one being a laptop to go jack to jack, but the other will just be fixed at the patch panel and just wired into the port you're testing. You should be able to get a link and connection thanks to MDI-X and you can simply set a static IP on each system so that they can talk to each other. Using this methodology, I would go through each port/jack combination and note your iperf results. Then you know where things stand. (I would also be VERY curious about your results since at some point I would love to try this test at my parent's place.) Once you have identified the marginal connections, you can check their termination or even just terminate the new unterminated wire even better since there are much better tools and ends now.

For 10Gb, if your Juniper does 10Gb on its SFP+, you might be able to get the Mikrotik 10GBase-T SFP+ if it will work and that will give you a 10Gb port panel-side. Or you could just get a nic for a machine and use the same Mikrotik SFP+ or just get a 10GBase-T card.

On the jack side I was thinking about the startech or sonnet thunderbolt 10GBase-T external nics. They aren't cheap, but are cheaper than having someone come in with a tester and run tests--plus you get to keep the nic if you want.

I'm thinking the whole 10Gb test setup should be under $500 and will give you some real-world insights based on the current 10GBase-T technology. :)
 

unmesh

Member
Apr 17, 2017
37
11
8
62
So I got my pair of USB 2.5G dongles and set about establishing a baseline using a short patch cable and two laptops, A with an i5-8250U 4 core 8 thread CPU and B with a i7-7600U 2 core 4 thread CPU. A is the server and B is the client and both laptops have the same latest Realtek drivers and show a link speed of 2.5Gbps.

iperf3 -c [server] gives about 1.8Gbps and iperf3 -c [server] -R gives 1.2Gbps over several runs. Another observation is that Windows Task Manager shows very low utilization on the server but almost 100% utilization on the client.

My desktop and servers unfortunately don't have USB-C ports to try out and I don't want to do any testing of the house wiring until I have confidence in the baseline.
 

Samir

Well-Known Member
Jul 21, 2017
1,534
507
113
46
Great work! I couldn't see full gigabit until I enabled some parallel streams
Code:
-P3
I would also try
Code:
-d
 

unmesh

Member
Apr 17, 2017
37
11
8
62
Didn't really make much difference. The reviews seem to say they get close to the spec so I'm mystified what is happening.

I also tried NTttcp and got 200-ish MB/s.

I'm going to see if I can buy a USB-C female to USB-A 3.1 cable so that I can try other computers.
 

Samir

Well-Known Member
Jul 21, 2017
1,534
507
113
46
iperf is a bit funny. Sometimes you do have to play with the options to get the rated bandwidth. Generally I've found this means running 3, 5, sometimes even 10 parallel.

I'd check the driver that's been installed. You might need to get the 'latest and greatest' straight from the manufacturer web site.
 

unmesh

Member
Apr 17, 2017
37
11
8
62
I am running the 'latest and greatest' driver straight from the manufacturer web site. This morning, I changed the power plan to high performance and put both laptops on AC but without a material difference. I also tried different numbers of parallel threads and noticed that the performance was unequal between threads although the sum was pretty similar.
 

unmesh

Member
Apr 17, 2017
37
11
8
62
Good news!

I made two bootable Ubuntu 20.04 flash drives and booted the laptops with those. Now iperf3 shows 2.35Gbps on the link in both directions and with 1/2/3/4 parallel threads.

Something is wonky with the Windows setup but I have my test rig :)
 

sleeper404

New Member
Jan 10, 2021
18
5
3
Both 2.5Gbps and 5Gbps multi-gig standards support cat5e so it shouldn't be an issue there at all. I've got a mix of cat5e/cat6 in my home and run a dual 10G adapter in my desktop. I've been too lazy to swap the last jumper out (cat5e) and it's links at 10G (cheap 10G-baseT SFPs) without an issue over about 15M.

2.35 seems perfectly reasonable with
 

unmesh

Member
Apr 17, 2017
37
11
8
62
Which 10G-BaseT pluggables are you using? I have SFP+ NICs and a switch currently connected using DAC cables but could use a couple of in-wall 10GbE connections over Cat5e if they work.

Thanks
 

sleeper404

New Member
Jan 10, 2021
18
5
3
I bought the slightly cheaper 6COM adapters at Amazon, but thinking back I would have tried these instead as they appear to support at least 2.5G as well from the comments section. My arista switch recognizes it as a optics module so it appears to present itself as a 10G-SR to bypass any copper module restrictions.

MultiGig-10Gbase-T Adapter at Amazon.
 

unmesh

Member
Apr 17, 2017
37
11
8
62
From what I've read, optical modules can only run at 10 and 1 so it would be interesting to see if the Arista will let you set it to 2.5 or 5 since it thinks it is an SR
 

sleeper404

New Member
Jan 10, 2021
18
5
3
I imagine the adapter itself negotiates with the switch as an emulated 10G-SR and then negotiates with the downstream device separately at 2.5/5/10G. I could be an issue if you were using these as part of the network infrastructure, but most protocols can handle setting an interface bandwidth command as the majority of links negotiate to 1G or 10G, but are less than such and it keeps IGP routing protocols from assuming a 150M MPLS link is actually 1G when calculating the metrics via DV/SPF.
 

unmesh

Member
Apr 17, 2017
37
11
8
62
I'm going to see if I can borrow a few modules from work and play with them though my Juniper switch is known to not support NbaseT