2.5 GE, more speed, more headaches

iljitsch

Ars Tribunus Angusticlavius
9,328
Subscriptor++
Way back in 2017 we expressed our collective desire for 2.5/5/10 Gbps networking.

Finally, a few days ago, I got a Mokerlink 5 x 2.5/1/0.1/0.01 Gbps + 1 x SFP switch for 55 euros. So a pretty good price for effectively five ports that range from 10 Mbps (which my 30-year-old Amigas with their 1990s NICs appreciate) to 2.5 Gbps, which is FINALLY a bit faster than 1 Gbps that my G4 PowerBook introduced me to MORE THAN TWO DECADES AGO.

I paid 100 euros extra for the upgrade from 1 to 10 Gbps on my Mac Mini back in 2020. I got a 2.5 GE USB dongle for my laptops a few years ago. So just hook up the 2.5 GE capable stuff, right?

Well, it's not that simple.

First, about NBASE-T: current 2.5+ Gbps UTP stuff will test the cabling and settle on a speed that will work well. Nearly all of my a-few-meter cat 5e cables would do 2.5 GE no trouble, but I actually have a nice somewhat longer red cable that only gave me 1 Gbps... Turns out that one is cat 5 without the e.

I previously outfitted my Synology DS218 NAS with a Cable Matters 10 - 2500 Mbps USB NIC and installed a driver. That didn't add a lot of speed, until I set the MTU to 9000 byte jumboframes. With that setting, I was able to saturate 2.5 Gbps, more or less.

My plan was to set up everything to use regular 1500-byte packets on the default (untagged) VLAN but then have a separate VLAN with a 9000-byte jumboframe VLAN. That worked just fine on the built-in Ethernet interface on my Mac Mini. Not so much on the Synology: you can't have additional VLAN interfaces that run on top of existing interfaces. So for the 2.5 GE interface, I was able to set a VLAN and jumboframes, but this was I lost access to the regular 1500-byte untagged network. So for that I had to use the built-in Ethernet interface.

Then I tried to use the Cable Matters / RealTek USB dongle to talk to the Synology. That was a clusterfuck, with much blame to be shared between Apple and RealTek. Back when I got one of these the first time, I had to install a driver to get access to all of the NIC's features. Of course that download page on realtek.com doesn't exist anymore. Well, at least Apple now shows the 2.5 GE capability for that interface. Well, if it shows the interface at all. But then no additional features such as jumboframes. Seriously, a 2500 Mbps interface in the 2020s is limited to 1500-byte packets? WTF??

Back in 2001 I was working on a project where we needed good 1 Gbps stuff. Remember, back then, 1000BASE-T was barely a thing. We had some vendors show us their stuff. I criticized one of them for not supporting jumboframes. A few weeks later they came back with 64000-byte jumboframe support.k This is all more than 20 years ago. How is it possible that between Apple and RealTek nobody bothered to support packets larger than the early 1980s limit of 1500 bytes!?!?

But it gets worse. On my 2016 MacBook Pro running whatever outdated MacOS version I can't remember the name of, I can't even manage VLAN interfaces in the System Preferences.

Still, it looks like even with 1500-mini-packets, I get pretty good performance (althout those numbers are all highly suspect) so perhaps the jumboframes are not needed after all.

In the meantime, the Mokerlink switch uses about 100 mW for a 10/100 Mbps port and something like 600 mW for a 1000 or 2500 Mbps port. My Mac Mini always keeps its Ethernet active even if the setting for that is disabled in the system preferences, but at least it drops down to a lower speed. So the Mokerlink switch uses virtually no extra power over my old 8 port 1 Gbps switch that only uses 2 W for all 8 ports. Interestingly, no "energy efficient ethernet" capability is reported.
 
Last edited:
FWIW, I'm sure use-cases for jumbo frames exist, but I haven't yet seen them really come into play with my own foray into 2.5GBE.

I have a pretty low-end setup (cheap unmanaged switches, etc) and get around 2.1Gbps moving files around via SMB to and from my NAS (same model as yours), with an MTU of 1500. Could I eke out a little more speed by messing with the MTU? Maybe, but I'm not convinced of the benefit/cost. Perhaps the fact that I'm running SSDs in my NAS factors in too.
 
Last edited:
  • Like
Reactions: Megalodon

steelghost

Ars Tribunus Angusticlavius
6,247
Subscriptor++
From what I've read on various reasonably tech-savvy forums (L1T, Proxmox, TrueNAS), modern computers are fast enough that the overhead of dealing with many small packets (as opposed to fewer large ones) doesn't necessarily have the same impact as it once did, especially for "low speed" network links like 2.5G.
 

Lord Evermore

Ars Tribunus Militum
2,570
Subscriptor++
FWIW, I'm sure use-cases for jumbo frames exist, but I haven't yet seen them really come into play with my own foray into 2.5GBE.

I have a pretty low-end setup (cheap unmanaged switches, etc) and get around 2.1Gbps moving files around via SMB to and from my NAS (same model as yours), with an MTU of 1500. Could I eke out a little more speed by messing with the MTU? Maybe, but I'm not convinced of the benefit/cost. Perhaps the fact that I'm running SSDs in my NAS factors in too.
High MTU helps with CPU usage on the device, so if you had a cripple-slow processor, or you were switching for an ISP backbone, jumbo frames might be helpful. As far as I know, that's pretty much it. Getting 0.1% higher throughput, maybe, since we already can just about max out gigabit, on a home network probably isn't worth the hassle, and would make your router have to constantly break up frames when sending traffic to the Internet, while the incoming (majority) would be standard 1500/1472. Probably some devices still don't support jumbo frames simply because of the extra support and design effort, and the number of people who would call complaining that their network is down because they enabled jumbo frames without making sure the other end supported them, or when they didn't see a massive speed improvement.

you can't have additional VLAN interfaces that run on top of existing interfaces
All VLAN interfaces are run on top of existing interfaces... 🙃

But yeah unreliable 2.5Gb would suck if I had a reason to use it. Maybe it's still at the stage that you need to ensure the same brand at every point. Which is stupid since it's actually been around a while. But then Intel had such a bitch of a time making the i225/i226v chipsets actually work, they could be to blame for it not being popular, too.
 

tucu

Ars Tribunus Angusticlavius
8,245
A few weeks ago one of my old GbE switches died and I reshuffle things by buying a 2.5GbE 8 port switch with a MaxLinear chipset. At the time it was €56 at Amazon after coupons. It is tiny and has been working fine with my 2.5GbE Realtek USB adapters and the Intel chipset in my mini PC. Power consumption seems too low for my smart plug to measure (IIRC I have measured 1W or 2W loads before).
FFbHqRa.png
 

iljitsch

Ars Tribunus Angusticlavius
9,328
Subscriptor++
2.5 GE at 1500 bytes with no optimizations means ~ 200k interrupts per second. No system can perform well in that situation.

Of course modern NICs come with all kinds of optimizations such as large segment offloading. But it sure seems like MacOS doesn't 100% take advantage of that.

But in any event, if the hardware supports larger packets (as it (not even going to cage with "virtually) always does these days), but the drivers can't be bothered to expose that to the OS, that is just BAD.

FYI, the Mokerlink switch says it supports 12k jumboframes, whatever that means. (12000 bytes? 12288 bytes?)

Conclusion: Apple sucks. RealTek sucks. People tried to work around that with open source RealTek NIC drivers but one of Apple's infuriating security efforts that no-one asked for breaks 9000-byte jumboframes. Apple SUCKS SUCKS SUCKS SUCKS. Obviously Linux does this stuff much better but Linux on the desktop SUCKs. Windows? SUCKS SUCKS SUCKS SUCKS SUCKS.

Why can't I have a moderately good operating system for my insanely good hardware in the 2020s?
 

iljitsch

Ars Tribunus Angusticlavius
9,328
Subscriptor++
All VLAN interfaces are run on top of existing interfaces... 🙃
On the Synology with the webinterface you get to set a VLAN ID but if you do that untagged packets are ignored. So you can't have both.

(Note that I wouldn't mind too much running the Synology with a jumboframe MTU on its main interface as TCP will negotiate the packet size with the MSS option. Not so much on my Macs though as this may break EDNS0 / DNSSEC, which often uses packets larger than 1500 bytes which my default gateway wouldn't be able to handle.)
 
Last edited:
I'm using jumbo frames on my 10G fiber network, and mixing 1500-byte units in doesn't seem to make any difference. Both the RPi 5 and the Mister FPGA unit can't go above 1500 byte MTUs, but they interoperate seamlessly with the other devices.

I don't think you need to segment your network, you can just run flat.

edit: I'm not using any Macs, though, just one Windows machine and a variety of Linux boxes.
 

Kyuu

Ars Praefectus
3,244
Subscriptor
But yeah unreliable 2.5Gb would suck if I had a reason to use it. Maybe it's still at the stage that you need to ensure the same brand at every point. Which is stupid since it's actually been around a while.
I can only speak IME, but I have a 2.5G PoE TrendNET switch feeding TP-Link Deco mesh WAPs which I occasionally connect to my laptop via an Asus 2.5G USB-to-Ethernet adapter. I don't think the issue is mixing brands, it all seems to work hunky dory (on a mix of cat5e/cat6 cabling with a mix of cat5e/cat6 rated terminations). There is a proliferation of no-name networking gear sporting 2.5G (and faster) ethernet ports which can definitely be very hit-and-miss.
 

Lord Evermore

Ars Tribunus Militum
2,570
Subscriptor++
I have some of that, and it works fine too. IMO 2.5G isn't any more problematic than 1G, the only downside is the strong likelihood that you don't really need it.
Man, if this had been available along with the high speed Internet services today, my roommate would have been saturating that link to a NAS almost continuously with porn. (He just downloaded it constantly and burned it to CDs which he collected in one of those huge folders. He didn't even watch most of it. And it was a particular genre.)
 
  • Haha
Reactions: CatBus

Demento

Ars Legatus Legionis
15,600
Subscriptor
For anyone running on the QNAP side of things, my research pretty much showed that the extra £10 (or whatever local currency) was worth it for QNAP's own 2.5Gbe PCIe adapter. I think any of the more recent (less than 10 years old) Intel-based NASen have at minimum a 2.0 x4 slot. I'm moving my network to 2.5G later this year. Realtek USB dongles and QTS play nice in theory, but people have all sorts of disconnect issues with them.
 
  • Like
Reactions: Lord Evermore

Lord Evermore

Ars Tribunus Militum
2,570
Subscriptor++
Actually optical drives are much too slow to keep up, with 1 x speed blu-ray at 100 Mbps IIRC. My drive is actually just USB 2.0 because that's fast enough.
CDs at 1X are 150 KILOBYTES per second! Blu-Ray 1X is 36Mbps. But what I meant was 2.5GbE going to a NAS, with multi-gig Internet, my old roommate would have been the archive for the Internet's porn (of a certain genre).
 

iljitsch

Ars Tribunus Angusticlavius
9,328
Subscriptor++
CDs at 1X are 150 KILOBYTES per second! Blu-Ray 1X is 36Mbps.
Oh right, I meant UHD (4K).

My Hitachi-LG BP55EB40 "Slim Portable Blu-ray WRITER" supports 2x and 4x for the Verbatim BD-Rs I have and I believe that is the maximum for that drive in general. So that would be 144 Mbps.
 
Last edited:

iljitsch

Ars Tribunus Angusticlavius
9,328
Subscriptor++
I'm using jumbo frames on my 10G fiber network, and mixing 1500-byte units in doesn't seem to make any difference. Both the RPi 5 and the Mister FPGA unit can't go above 1500 byte MTUs, but they interoperate seamlessly with the other devices.

I don't think you need to segment your network, you can just run flat.
I wouldn't do it if your router can't do the jumboframes. Because then you can't talk to others elsewhere who use larger packet sizes.

BTW, my Raspberry Pi 400 can also not do packet sizes larger than 1500 (although the 2006 update of IEEE 802.3 now specifies 2000-byte frames, and I'm pretty sure all even remotely recent hardware supports that). But that Cable Matters dongle will accept an MTU of 16 kilobytes -18 (Ethernet overhead) -4 (VLAN header size) but that doesn't work. 9000 bytes no problem, though, and transfers are a bit faster, but nothing anyone would lose any sleep over.
 
I wouldn't do it if your router can't do the jumboframes. Because then you can't talk to others elsewhere who use larger packet sizes.
It's a Linux mini-PC, so it handles that fine. The internal MTU is 9000, the other interfaces are 1500. I hadn't even considered trying the external interface at 9K, that seemed like looking for trouble. Keeping jumbo frames local seemed like it was the best idea.

Hmm, now I'm curious whether my ISP will even deliver 9K frames. I'll have to experiment with that.

9000 bytes no problem, though, and transfers are a bit faster, but nothing anyone would lose any sleep over.
That might make a noticeable difference with the weaker CPU in the Pi series, but with the I/O being so limited, I'd rather just suffer along on the crappy built-in gigabit on the 5, and save the USB bandwidth for storage. It's not like it has any trouble talking to the large-MTU machines, so it's not really a problem I need to solve.
 

Megalodon

Ars Legatus Legionis
36,693
Subscriptor
From what I've read on various reasonably tech-savvy forums (L1T, Proxmox, TrueNAS), modern computers are fast enough that the overhead of dealing with many small packets (as opposed to fewer large ones) doesn't necessarily have the same impact as it once did, especially for "low speed" network links like 2.5G.

It's not the speed it's more the offload features in the chips that avoids work for the OS network stack.

2.5 GE at 1500 bytes with no optimizations means ~ 200k interrupts per second. No system can perform well in that situation.

Of course modern NICs come with all kinds of optimizations such as large segment offloading. But it sure seems like MacOS doesn't 100% take advantage of that.

Strongly suspect the regression is elsewhere in the stack.

Conclusion: Apple sucks. RealTek sucks.

This isn't realtek's fault.
 

iljitsch

Ars Tribunus Angusticlavius
9,328
Subscriptor++
Yeah, buy stuff, it stops working a few years later.

If I have an old computer that needs an old driver that was available several year ago, removing that driver from your downloads SUCKS SUPER BAD. I don't want to deal with companies that thing that's reasonable.

The stupid thing is that there was a reasonable driver years ago but not anymore for something recent (if not 100% up-to-date, as Apple's hardware quality vastly outpaces its software support for older hardware).

(In this case, an Intel Mac Mini running MacOS Sonoma. Which still kernel panics multiple times a week. Why did I upgrade???)
 

w00key

Ars Tribunus Angusticlavius
9,153
Subscriptor
Your trouble seems to be new hardware x old OS. That rarely ends up well, like that Windows install I Frankensteined, old LTSC build, new hardware => all the drivers I manually downloaded require a newer Windows, officially. The Intel driver assist tool and Windows Update are like 🤷‍♂️, no drivers for you sir.

Sure it kinda works after just clicking setup.exe on the GPU, network driver etc but they all officially require something like 22H2+. Some drivers refused to install leaving yellow ?'s in device manager. Meh, works well enough, I'll nuke it later.
 
  • Like
Reactions: Gregg S

iljitsch

Ars Tribunus Angusticlavius
9,328
Subscriptor++
Your trouble seems to be new hardware x old OS. That rarely ends up well
How does this statement make sense? Software is supposed to get better, not worse.

In this case my old OS doesn't have a good built-in driver, but I could download a reasonably good one from the chip maker. Now the built-in driver is mediocre but nothing from the chip maker.

So whatever OS you're running, nothing really good.

I currently can't add VLAN interfaces in MacOS. Who are the idiots in charge of this OS and why are they getting a paycheck??

I was already highly unhappy with Apple, but with their new US administration accommodation policies, I think it's time to stop buying new Apple stuff.
 

w00key

Ars Tribunus Angusticlavius
9,153
Subscriptor
In this case my old OS doesn't have a good built-in driver, but I could download a reasonably good one from the chip maker. Now the built-in driver is mediocre but nothing from the chip maker.
Newer drivers are often made for newer OS. That's my point. Intel didn't build GPU drivers for the 12-14th gen / Core Ultra for a 2020 Windows 10 release, it requires Win 10 22H2 and newer. I doubt anyone making new hardware available now will spend extra time and code against an old 2020-era Windows DDK instead of just targeting the recent ones, like 22H2 and up.

You see that with open source too, want new OSS drivers? Get a new kernel. You will not run Ubuntu 18.04 with its 4.15 kernel on new builds with full hardware support, if it boots at all. To help with that Ubuntu has Hardware Enablement releases like 18.04.5 LTS (HWE) that bumps the kernel to 5.4, but that's it. No 5.15 / 6.x for you, OS is too old.


Android / iOS apps too - we are on a treadmill to upgrade / target the latest version within 1 year of release, I'm not going to give it too much thought keeping it compatible with ancient devices. Last 3 or 4 releases is plenty. If they added a feature worth upgrading for, or if it makes my life easier by requiring less "if version is this" compatibility code, I may drop support early.
 

iljitsch

Ars Tribunus Angusticlavius
9,328
Subscriptor++
Android / iOS apps too - we are on a treadmill to upgrade / target the latest version within 1 year of release, I'm not going to give it too much thought keeping it compatible
Please let me know your products so I can make sure to never use them.

I got an iPhone SE a little over a year ago.This was the last model with a smaller but square-cornered un-notched screen. I now worry how long that will be supported properly.

I'm guessing this is my last iPhone. I really hate the direction all of this stuff is going.

(Did I mention my 20-year-old Nokia phone still works just as well as ever WITH ITS ORIGINAL BATTERY?)
 

w00key

Ars Tribunus Angusticlavius
9,153
Subscriptor
Please let me know your products so I can make sure to never use them.

I got an iPhone SE a little over a year ago.This was the last model with a smaller but square-cornered un-notched screen. I now worry how long that will be supported properly.

I'm guessing this is my last iPhone. I really hate the direction all of this stuff is going.
You're not paying for maintenance and development, the ones paying my bills are.

But on iOS support should last long enough. If it's the latest mini released in 2022, I expect you can install the newest iOS until at least 2027. And then you still have a few years of grace period where app will still work. As I said, we usually target last 3 major versions so that's at least 2030.

Android is much worse, you often only get 2 or 3 major releases and that's counting from release date, so a budget Samsung from 2022 is already out of support and apps may stop updating after a few years. You can still use an old version, just not the last one.

(Did I mention my 20-year-old Nokia phone still works just as well as ever WITH ITS ORIGINAL BATTERY?)
Sure, as a clock. I doubt any provider still issues SIM cards compatible with it or run a classic GSM network. Even 2G (EDGE) and 3G (UMTS) has been phased out here on most virtual providers and 1/3 major national networks.

There are still some networks running in compatibility mode for smart meters and alarm units but this service will also end in the next few years. Utilities pay a lot of keep the last GPRS network running. As a consumer, you can't purchase that service anymore. After that, it's LTE / 5G only.


Old things that keep working are ones with no network dependency. A phone won't work forever. Nor do apps with an internet connection. When Let's Encrypt switched root certificate and chain of trust, older devices stopped trusting it as SSL/TLS anchor because it doesn't know the new root. So gg, these devices suddenly can't talk to a major chunk of the Internet.

Only pure offline devices last forever. Like a Super Nintendo. Anything newer, like a DS with wifi, will lack Internet hosted features after a while. But with cartridges, not eShop downloads, most games should still work.
 
  • Like
Reactions: Gregg S

iljitsch

Ars Tribunus Angusticlavius
9,328
Subscriptor++
(About my Nokia 6310i still working)
Sure, as a clock. I doubt any provider still issues SIM cards compatible with it or run a classic GSM network.
Apparently not, as a backup SIM that I got a few years ago didn't work.

Even 2G (EDGE) and 3G (UMTS) has been phased out here on most virtual providers and 1/3 major national networks.

There are still some networks running in compatibility mode for smart meters and alarm units but this service will also end in the next few years.
Sure.

But here in NL it's still running. And my main SIM (that I use with my 2022 iPhone SE) is old enough that the Nokia recognizes it and I was able to make calls.

Only pure offline devices last forever. Like a Super Nintendo. Anything newer, like a DS with wifi, will lack Internet hosted features after a while. But with cartridges, not eShop downloads, most games should still work.
My Amiga computers from the early 1990s still work. It's really nice that I can access my NAS on them, although not even maxing out 10 Mbps Ethernet. The software to be able to do that was created more than a decade after the last Amiga was sold.

Then again, the Safari that was last updated in 2021 on my 2013 MacBook Pro is no longer capable of handling more and more websites.
 

Katagi

Ars Centurion
327
Subscriptor
And why can't I download the older driver anymore?
What driver are you looking for? The open source driver you linked was for RTL8125. Does the driver here work for you? This is for MacOS 10.7 on an Intel-based Mac computer.

You mentioned that there used to be an older driver that worked. It is possibly still available somewhere if not at the official source. The internet archive is a good first search for old software.
 

iljitsch

Ars Tribunus Angusticlavius
9,328
Subscriptor++
(About my Nokia 6310i still working)
Some more data:

I got two new SIMs that are clones so I can use the same plan on two phones, but not at the same time. So these are brand new SIMs and they work on the Nokia. These are from KPN NL.

Also, although I did have trouble with a ~ 3 year old Vodafone NL SIM in that Nokia previously, it now works without issue.

So apparently not too much new tech in SIMs over the past quarter century, although of course now we have eSIMs. What's nice about my EU iPhone SE 3G is that it still has a physical SIM slot, which you can use along with an eSIM, you can also use two eSIMs at the same time.
 

torp

Ars Praefectus
3,425
Subscriptor
Oh, speaking of 2.5 Gbps... can anyone recommend a low power 16 port switch that has like 4 (or a few more, i don't mind) >= 2.5 Gbps ports? The rest can be 1 Gbps.

Don't need fancy management features, I have only a faint idea what a VLAN is. I'm just thinking of using the > 1Gbps speed on the few newer devices that I have that support it.

I'm assuming that getting a 4-5 port 2.5 switch and plugging my old 16 port in it would create a bottleneck when the devices on different switches talk to each other.
 

ERIFNOMI

Ars Legatus Legionis
18,134
Oh, speaking of 2.5 Gbps... can anyone recommend a low power 16 port switch that has like 4 (or a few more, i don't mind) >= 2.5 Gbps ports? The rest can be 1 Gbps.

Don't need fancy management features, I have only a faint idea what a VLAN is. I'm just thinking of using the > 1Gbps speed on the few newer devices that I have that support it.

I'm assuming that getting a 4-5 port 2.5 switch and plugging my old 16 port in it would create a bottleneck when the devices on different switches talk to each other.
This is going to sound stupid, but it's only a bottleneck if it's a bottleneck. If that second switch has half a dozen IP cams streaming at a few Mbps each back to an NVR on the main switch, a 1Gbps link is barely being used. If you've got two data hungry users on the second switch both trying to pull down large files from a file server on the first switch simultaneously, yeah, they're going to share 1Gbps between them. So if you have a handful of pretty low throughout devices (and you probably do in a home environment, even video streaming doesn't need that big of a pipe), you can drop those on the second switch.
 

Demento

Ars Legatus Legionis
15,600
Subscriptor
This is going to sound stupid, but it's only a bottleneck if it's a bottleneck. If that second switch has half a dozen IP cams streaming at a few Mbps each back to an NVR on the main switch, a 1Gbps link is barely being used. If you've got two data hungry users on the second switch both trying to pull down large files from a file server on the first switch simultaneously, yeah, they're going to share 1Gbps between them. So if you have a handful of pretty low throughout devices (and you probably do in a home environment, even video streaming doesn't need that big of a pipe), you can drop those on the second switch.
And I can't comment on greater than 8+2, but there are plenty of those on AE that do the job just fine*.

*caveat that I haven't used the 2 10Gb SFP ports.
 
Oh, speaking of 2.5 Gbps... can anyone recommend a low power 16 port switch that has like 4 (or a few more, i don't mind) >= 2.5 Gbps ports? The rest can be 1 Gbps.

Don't need fancy management features, I have only a faint idea what a VLAN is. I'm just thinking of using the > 1Gbps speed on the few newer devices that I have that support it.

I'm assuming that getting a 4-5 port 2.5 switch and plugging my old 16 port in it would create a bottleneck when the devices on different switches talk to each other.
Your use case doesn't sound too demanding, so I'm throwing this idea out there. 16 port switches tend to be more upmarket, because your average home user doesn't need that many ports. So, one thing you could do is just link together two 8-port switches. You're effectively making a single 14-port switch (one port on each switch dedicated to the link), with the downstream switch limited to 1/2.5Gbps total traffic over that link. But that might be fine.

A slightly more upscale variation on the same theme is to pick up two of something like this:
https://www.servethehome.com/gigaplus-gp-s25-0802-8-port-2-5gbe-and-2-port-10g-review/

You can then link them on the 10Gbps ports and you end up with 16 2.5Gbps ports (and two bonus 10Gbps ports), and more bandwidth than you're likely to ever run into problems with on a basic (for Ars Technica!) home network. Power utilization of two of these devices versus one larger device is likely negligible. I use these exact switches exactly like this (linked using a DAC cable) with no issues. They're dumb unmanaged fanless switches, but they pass VLAN traffic if that's ever going to matter to you.
 
Last edited:

BigLan

Ars Tribunus Angusticlavius
7,882
Chiming in here as I just did a 2.5gb upgrade last night. My 24 port gig switch seems to have died so I got one of the cheap 8x2.5gb + 2xSFP switches from AliExpress and a few pcie and USB adapters for the PCs in the house. They're all Realtek based, popped them in and put the switch in the wiring closet and everything just worked - even the reused Cat5e cable that had been used for telephone lines 20+ years ago.

I've currently got an 8 port gig switch hanging off the 2.5gb switch handling back haul for my mesh nodes and other stuff that don't need the higher speed (Xboxes, tvs etc) but have similar concerns about that becoming a bottleneck at some point. I can't see any 8port gig + sfp switches on Ali, so maybe just getting another of the 2.5gb switches is in my future - they're not that much more expensive anyway (well, until tariffs/de minimus kicks in.) There are loads of cheap 100mbit poe switches with SFP, but I guess the gig versions can't be built cheaper than the 2.5gb.

Now I'm wondering if I need to run fiber from my media server to the switch, just when I thought I was done crawling around in the attic.
 
  • Like
Reactions: Demento

Lord Evermore

Ars Tribunus Militum
2,570
Subscriptor++
even the reused Cat5e cable that had been used for telephone lines 20+ years ago
2.5GbE and 5GbE are explicitly intended to be drop-in upgrades for homes and small businesses that won't require a cabling upgrade, both intended to work on Cat5e. (5Gb does need Cat6 for longer distances, but not Cat6a like 10Gb, and many places future-proofed a bit and already have Cat6 if they've had wiring installed in the last several years so for many users it is still a drop-in upgrade.)
 

ERIFNOMI

Ars Legatus Legionis
18,134
2.5GbE and 5GbE are explicitly intended to be drop-in upgrades for homes and small businesses that won't require a cabling upgrade, both intended to work on Cat5e. (5Gb does need Cat6 for longer distances, but not Cat6a like 10Gb, and many places future-proofed a bit and already have Cat6 if they've had wiring installed in the last several years so for many users it is still a drop-in upgrade.)
10GBASE-T allows for Cat6 for runs up to 55m by spec. In a home environment, Cat6 should be fine for 10Gbps. 55m is a long run, even if you have to go weird routes to get there, and that's just "spec." You can probably get away with even longer runs in a relatively quiet home environment where you have a few runs, not 50-100 (I recognize the amount of Cat6 in my house is very, very atypical).