Yup, decade later and the market for 10G consumer switches is nearly non-existent. Everything is stuck at 1G for everything.
Were I to guess, I'd say it's probably due to internet speeds in the US taking forever to get up to 1Gbps. There's little reason to push out consumer switches faster than 1Gbps when the inbound internet connection only goes up 100Mbps.
I'd say it's between that and the vast majority of home devices switching to Wi-Fi anyways.
To add a 3rd more minor reason: for the prosumers DACs/fiber is faster, cheaper, and uses less power. The only reason to go to 10GBASE-T is for cases with existing cabling where 2.5G/5G might make a lot more sense if you don't absolutely need 10G. For everything else go 10G/25G passive DAC or fiber.
10G also isn't compatible with most of the lower-CAT cabling that's already run in the walls. Most businesses aren't going to tear it all out to start over.
802.11ax can go >10gbps without needing any fixed infrastructure. 10GBASE-T is basically abandonware now, at least for consumer applications.
For servers, fiber is even more scalable (25/50/100/400G), and about as easy to handle.
Yeah cabling is what will usually make 2.5G/5G make a lot more sense if you don't absolutely need the full bandwidth, those speeds give you options for >1G without needing to rip your cabling plant out.
Not quite >10 Gbps on 802.11ax though - standards compliant gear has a max theoretical cap of 9.61 Gbps half duplex using 8 spatial streams and 2.4 Gbps half duplex if one end is any current client chip which is limited to 2 spatial streams. Realistic numbers a little lower, I can get ~2 Gbps goodput on an AX210 in the same room as the router on an empty 160 MHz channel. Still far more than any home consumer will need but it's a bit of a ways off of obsoleting wired 10 Gbps full duplex throughput, especially with the cost of 2 8x8 APs.
I agree with your guess. But we're starting to see more than 1gbps speeds, however, not everyone will subscribe to them.
Xfinity sells 1.2gbps (used to be 1gbps) and 3gbps (used to be 2gbps) to homes, but the upload speeds are horrible. At&t with its fiber offerings just announced 5gbps speeds, but they're very expensive and have horrible practices that prevent you from using anything but their router.
I've got AT&T's 1Gbps service and you can bypass their router. You just have to log into the web UI and set it up there. At least I was able to do it with mine.
That's not true bypassing, you're still required to keep their router connected. There is another bypass method I used that allowed me to disconnect the router but it was hacky and difficult to execute.
The main reason we try to bypass AT&T's router is because their NAT table is capped to accept 2048 entries and starts dropping sessions after it reaches the limit. This is why I stated that the web UI method you mentioned is not a true bypass.
The gist is that you'll have two ethernet ports dedicated to AT&T, one for the ONT, and one for the gateway router (which needs to have the gateway's mac address). The pfSense interfaces have to be setup a specific way. There is also a script that has to run while pfSense boots. Once it boots, you can pretty much turn off the gateway and put it away. The only time you'd need to reconnect it is if your ONT loses power.
It will help, tysm! My hardware is a bit different (SFP module so no separate ONT and gateway) but it should be possible to cook something similar given the detail provided (e.g. VLAN0)
I'm thinking about using JTAG or something to extract the key, because I'd much prefer having everything handled by my own server.
I'd still be happy to pay AT&T the monthly rental for their modem and for the privilege of leaving me alone :)
Any new/recent gig or higher installs will use the BGW320 which, apart from not tanking performance in passthrough mode, will do 8192 sessions. Not exactly an amazing cap but also not "door frame at head height" level anymore.
The xfinity 3 gbps service (actually 3+1) is symmetric. And it’s not shared at the neighborhood level—you’ve got a dedicated 10g port on a Comcast switch at your head end. Of course it costs $300/mo.
Bell Aliant (Atlantic Canada) offers 2.5G down with an SFP+ stick. Upload is limited to 940 Mbps as the optic uplink laser isn’t strong enough to support higher speeds (my understanding). I’m too cheap to pay for anything more than 100/100 service, so an Edgerouter X SFP is sufficient for my bypass needs.
Bell in other parts of Canada is starting to offer XGPON service (up to 10 Gbps) with a unit that has an integrated ONT, so bypass gets trickier.
What I also noticed is they made both 1Gbit and 2.5Gbit Ethernet compatible with Category 5 cables from the 1990s. Even the difference between Cat5 and Cat5e are quite minor.
I thought part of the draw of io_uring was that, at present, having one process saturate the network was challenging if not impossible.
My read is that we are in a bubble where some of the 8 Fallacies of Distributed Systems don't have all of their teeth because, while none of the Fallacies are specifically about noisy neighbors, several of them are decidedly informed by them. If I can't outshout you on the network, then you can be forgiven for ignoring the consequences of network congestion. If we figure out how to saturate a 10GB link between two threads running on different machines, then there will suddenly be a lot of remakes of old lectures.
I would quite happily upgrade to 10GBASE-T now on my home network but alas the prices are still a bit too much for a reasonable consumer upgrade. In general I am a bit disappointed with consumer network kit, we could do with some openWRT/DDWRT/Tomato Wifi 6E capable routers able to do CAKE QoS on gbit fibre. Rather than having to hack a raspberry pi 4 together with a consumer router as an access point for the wifi.
What I want is 10GBASE-T and wifi 6E and the performance to route 1gbit/s with QoS and I would like it for something more like $200 or less. $250 for a couple of 2.5gbit/s ports on a switch is just not worth it.
I got a microtik 4 port 10Gb switch and some cheap eBay cards to go with the 10Gbe port on my thunderbolt dock - works well for VMs and NAS stuff. Everything else is gigabit or wifi.
If I go to Mikrotik's Switches section and filter by "10 Gigabit combo interface: 1 - 4", there's only one $600 product (CRS312-4C+8XG-RM). Is that the one you're using?
It's an ethernet switch with one 1Gbase-T port and 4x SFP+ ports that can be any media you want the proper transceiver. Including 10Gbase-T if you really want it for some reason.
Yep, and I have two DAC, one fiber, and one twisted pair SFPs in it. Works well and DACs aren’t exceptionally costly. The fiber goes to a 10g uplink on the old core switch I got on eBay.
> 10GBASE-T and wifi 6E and the performance to route 1gbit/s with QoS and I would like it for something more like $200
In one device? You want too much for too less, each one separately costs ~$200. And there is no demand for such devices yet which is the reason you can't find it.
2.5G and 5G Ethernet are relatively cheap. I recommend them for new installs.
I had a 10GBASE-T network for a while but returned to 1G and haven’t really missed it. It only made a difference during sustained reads/writes from the NAS, and my workflow almost never gets blocked on that. The switches and even Ethernet cards are power hungry and, unless you’re careful about switch selection, can be noisy. It just wasn’t worth it for me.
It's not what I saw a bit more than a year ago when I changed my PC. A PCI express card was about 100€, about the same for a switch IIRC. It was about 500€ total for what I wanted to do, not worth it for me.
The same hardware just Ethernet would have cost 5 or 6 times less!
Motherboard ports are rapidly becoming 2.5G. Adapter cards are $35. Switches about $100, depending upon what you want (getting to a managed switch with a lot of multigig ports is $400, but then you also tend to get PoE).
You're right, I just checked on my Newegg equivalent and the local Amazon, one has PCI express cards for 35€ and the other has switches for 135€. Too bad the price for switches didn't go down, I have a hard time replacing 15€ switches for 135€ one just for faster transfers.
Every time I've looked at upgrading to anything better than 1G Ethernet I can't justify the cost for the performance. Only my NAS can come close to saturating 1G but I rarely do that. Everything else in the house is more than happy with 1G and WiFi. I can't even get my WiFi 6 AP to get higher than about 500Mbps.
Even if I wanted to upgrade it means replacing/upgrading my switch, router, and wiring in some cases. Going from Fast Ethernet to 1G was a no brainer and relatively inexpensive. For home use 10G is just too rich for my blood and won't get me much better actual network performance.
In that sort of situation, a high speed switch and one high speed network card (in the NAS) would seem to be the biggest bang for the buck.
Switches with one, maybe two high speed uplink ports used to be quite common, allowing every machine on the switch to get around 50% utilization of their network connection before congestion set in.
My NAS has two 1G ports that I set up as a bonded interface. So it's got 2Gb of capacity which is more than enough capacity for any other device on the network. Even upgrading to a 2.5G or 5G switch isn't going to do much for my network capacity and just cost me a bit of money.
10GBASE-T has rather insane computational requirements: way more work per bit compared to predecessors. And 10x the bits. A pretty sizeable chunk of silicon is needed to do all of it. Hence high power consumption and price-tag. Even now, 15 years later. I guess standard's authors had a little bit too much faith in Moore's law.
> LDPC FEC coding [2048,1723], with the parity check matrix construction based on a generalized Reed–Solomon [32,2,31] code over GF(26).
There was a point in time circa 2014-16 where it would make sense in some datacenter scenarios. IIRC, top of rack switching and copper cable was cheaper than running fiber.
Pretty sure that’s not true anymore, and with workloads moving to cloud, there’s probably no investment/scale.
For Tor, DAC was cheaper and more accessible. By 2016, the prices for 10Gbase-T and SFP+ switches was often close to the same, but SFP+ gave significant flexibility while 10Gbase-T meant you had to architect specifically for it.
With SFP+ hardware we had coax DAC for cheap connections, precut cheap fiber with hardwired transceivers, and of course expensive transceivers for longer range connections on fiber, and we could mix and match.
A 10Gbase-T device meant we had to build the entire segment for new type of wiring, ensuring right NICs and switches, because mixing 10Gbase-T with SFP+ always required expensive transceivers (they were also not trivial to acquire in 2016, speaking from experience trying to unfuck a purchase of Nexus 3k in 10Gbase-T variant when all servers had SFP+)
But then if you had 10G you wouldn't be running full rate of the line so you'd need to switch to an array of SSDs but then you wouldn't be running full rate of the line so you'd need to switch to 100G ethernet but then you would be limited by CPU/RAM performance so you'd need to buy a xeon/epyc but then...
Or just buy a modern NAS, which will already have multiple NVMe drives, arrays of SSDs, and a decent number of spinny drives. outside of low end consumer kit, saturating 10G is easy with any modern equipment. Even 5G is easy with my 8 year old NAS.
I upgraded my home to 10Gbps networking once I got 2Gbps fibre. I use fiber or DAC for most links unless I have to run through my existing Ethernet wiring. My primary switch is just a bunch of SPF+ ports into which I can stick a bunch of different type of connections in.
One decade later: 1 GbE means hosts are capped at 4 to 6 simultaneous 4k bluray streams, or loading a single 400GB game per hour.
Still fast enough for most consumers. Outside of a home lab, only maybe a NAS and one or two hosts transferring lots of bulk data there need bigger pipes, and even then 10GbE usually means "throttled by disk speeds or computational protocol overhead" (e.g. Samba on a slow CPU; anecdata: I see about 200MB/s over SMB on my 10GbE fiber link, but invested zero time in optimizing things). And in professional environments I'd be surprised to see 10GBASE-T instead of an equivalent fiber (less power draw = less heat, also better scaling; only drawback might be when replacing legacy copper cabling).
Looking at the storage on my PS5, the largest game on it is the preload for "Horizon: Forbidden West" which goes live next Friday. It's 88GB. It's an open world RPG with several dozen hours of gameplay.
"Returnal", about a year older and a much shorter game, is the next largest at 60 GB.
Toward the bottom of the list are "Maneater" and "Life Is Strange: True Colors", both clocking in at about 15GB and as many hours of gameplay.
Having said that, your net connect isn't going to help with games. I have Starlink, which generally sits in the 150-250Mbit range. That should have taken about an hour to download 88GB. It took 7. The CDNs for game delivery aren't exactly speedy. (And that's pretty common, in my experience. Even in my last place when I had gigabit fiber, I generally left the console downloading overnight.)
I'm not in the loop here, most of the games I've installed from steam are in the 1GB to 20GB range, with a few outliers up to 80GB (I guess). IIRC some recent AAA games are in the 200GB range due to HQ textures. So I think 400GB games are realistic within the next decade. But honestly, it was just a random example. Just s/single 400GB game/400GB of whatever/ ;-)
Fiber SFP+ are much less energy-intensive, the cables are thinner and more manageable, and have longer distance as well as better reliability. It's time to bite the bullet and go optical.
It depends on what you are doing. For office/lab space, twisted-pair (Cat6 or better) is really compelling. You can wire an office/lab space and support both 1- and 10-Gigabit connection.
That being said, it's a good idea to do one of two things:
1. Run some fiber to a centrally-located faceplate in each space, with MT-RJ jacks (or something else, like LC).
2. Run a conduit (an inner duct) to each office/lab space, with a pull tape, so that fiber can be easily pulled through in the future.
I personally prefer #2, for reasons like this: We have a few buildings with fiber run into offices/labs, but the fiber run was multimode. That's fine if you want to connect to a switch in the floor or building network closet, but it's not going to be compatible with the inter-building fiber, which is all singlemode.
So... Apple shipped a gigabit Ethernet desktop in 2000 and a gigabit Ethernet laptop in 2001.
20+ years later, laptops have 40Gb Thunderbolt and 10Gb USB, but Thunderbolt or USB to 10Gb Ethernet interfaces are still expensive, though 5Gb and 2.5Gb seem to be becoming cheaper and more common. And even a basic, unmanaged 5-port (e.g. TP-Link) 10G/5G/2.5Gb Ethernet switch is still something like $300.
That was an example, which remains so for the reasons you outlined. My point was, if we designed things to fit within smaller constraints than we actually have, we would be able to do more with the same resources. We wouldn't need 8GB to run two instances of an app if it was designed to fit in 2GB instead of 4.
Still takes awhile to process 10TB over Gig, and it’s easy to end up with that quickly for everything from surveillance footage, 4K video, any sizable number of high resolution photos, etc.
As 5Gbps Ethernet is somehow only included in 10Gbps products. I wish as bare minimum that 2.5Gbps Ethernet to completely replace 1Gbps Ethernet. But with current chip supply issue it will be another few years before we see wide adoption.
It is strange when some Nordic countries are already getting affordable 2.5 to 10Gbps Internet.
69 comments
[ 3.7 ms ] story [ 131 ms ] threadWere I to guess, I'd say it's probably due to internet speeds in the US taking forever to get up to 1Gbps. There's little reason to push out consumer switches faster than 1Gbps when the inbound internet connection only goes up 100Mbps.
To add a 3rd more minor reason: for the prosumers DACs/fiber is faster, cheaper, and uses less power. The only reason to go to 10GBASE-T is for cases with existing cabling where 2.5G/5G might make a lot more sense if you don't absolutely need 10G. For everything else go 10G/25G passive DAC or fiber.
802.11ax can go >10gbps without needing any fixed infrastructure. 10GBASE-T is basically abandonware now, at least for consumer applications.
For servers, fiber is even more scalable (25/50/100/400G), and about as easy to handle.
Not quite >10 Gbps on 802.11ax though - standards compliant gear has a max theoretical cap of 9.61 Gbps half duplex using 8 spatial streams and 2.4 Gbps half duplex if one end is any current client chip which is limited to 2 spatial streams. Realistic numbers a little lower, I can get ~2 Gbps goodput on an AX210 in the same room as the router on an empty 160 MHz channel. Still far more than any home consumer will need but it's a bit of a ways off of obsoleting wired 10 Gbps full duplex throughput, especially with the cost of 2 8x8 APs.
Xfinity sells 1.2gbps (used to be 1gbps) and 3gbps (used to be 2gbps) to homes, but the upload speeds are horrible. At&t with its fiber offerings just announced 5gbps speeds, but they're very expensive and have horrible practices that prevent you from using anything but their router.
The main reason we try to bypass AT&T's router is because their NAT table is capped to accept 2048 entries and starts dropping sessions after it reaches the limit. This is why I stated that the web UI method you mentioned is not a true bypass.
The gist is that you'll have two ethernet ports dedicated to AT&T, one for the ONT, and one for the gateway router (which needs to have the gateway's mac address). The pfSense interfaces have to be setup a specific way. There is also a script that has to run while pfSense boots. Once it boots, you can pretty much turn off the gateway and put it away. The only time you'd need to reconnect it is if your ONT loses power.
I hope that helps.
I'm thinking about using JTAG or something to extract the key, because I'd much prefer having everything handled by my own server.
I'd still be happy to pay AT&T the monthly rental for their modem and for the privilege of leaving me alone :)
https://broadbandmap.fcc.gov/#/provider-detail?version=dec20...
Bell in other parts of Canada is starting to offer XGPON service (up to 10 Gbps) with a unit that has an integrated ONT, so bypass gets trickier.
My read is that we are in a bubble where some of the 8 Fallacies of Distributed Systems don't have all of their teeth because, while none of the Fallacies are specifically about noisy neighbors, several of them are decidedly informed by them. If I can't outshout you on the network, then you can be forgiven for ignoring the consequences of network congestion. If we figure out how to saturate a 10GB link between two threads running on different machines, then there will suddenly be a lot of remakes of old lectures.
What I want is 10GBASE-T and wifi 6E and the performance to route 1gbit/s with QoS and I would like it for something more like $200 or less. $250 for a couple of 2.5gbit/s ports on a switch is just not worth it.
CSS610-8G-2S+IN Eight 1G Ethernet ports and two SFP+ ports for 10G fiber connectivity. $99.00
In one device? You want too much for too less, each one separately costs ~$200. And there is no demand for such devices yet which is the reason you can't find it.
I had a 10GBASE-T network for a while but returned to 1G and haven’t really missed it. It only made a difference during sustained reads/writes from the NAS, and my workflow almost never gets blocked on that. The switches and even Ethernet cards are power hungry and, unless you’re careful about switch selection, can be noisy. It just wasn’t worth it for me.
It's not what I saw a bit more than a year ago when I changed my PC. A PCI express card was about 100€, about the same for a switch IIRC. It was about 500€ total for what I wanted to do, not worth it for me.
The same hardware just Ethernet would have cost 5 or 6 times less!
Even if I wanted to upgrade it means replacing/upgrading my switch, router, and wiring in some cases. Going from Fast Ethernet to 1G was a no brainer and relatively inexpensive. For home use 10G is just too rich for my blood and won't get me much better actual network performance.
Switches with one, maybe two high speed uplink ports used to be quite common, allowing every machine on the switch to get around 50% utilization of their network connection before congestion set in.
The designs are just a generation ahead in everything, and they are really pulling close to 1g theoretical maximum.
> LDPC FEC coding [2048,1723], with the parity check matrix construction based on a generalized Reed–Solomon [32,2,31] code over GF(26).
> Tomlinson-Harashima precoding (THP)
[0] https://en.wikipedia.org/wiki/10GBASE-T#10GBASE-T
Pretty sure that’s not true anymore, and with workloads moving to cloud, there’s probably no investment/scale.
With SFP+ hardware we had coax DAC for cheap connections, precut cheap fiber with hardwired transceivers, and of course expensive transceivers for longer range connections on fiber, and we could mix and match.
A 10Gbase-T device meant we had to build the entire segment for new type of wiring, ensuring right NICs and switches, because mixing 10Gbase-T with SFP+ always required expensive transceivers (they were also not trivial to acquire in 2016, speaking from experience trying to unfuck a purchase of Nexus 3k in 10Gbase-T variant when all servers had SFP+)
Upgrading clients to 10 ends up being quite expensive though so will have a mixed network. Plus some form factors don't have pcie slots available
For extremely large files it makes a large difference but otherwise I haven’t seen any difference at all.
The network card is more expensive, it generates more heat and uses more electricity. I pretty much regret spending the money to upgrade.
"Returnal", about a year older and a much shorter game, is the next largest at 60 GB.
Toward the bottom of the list are "Maneater" and "Life Is Strange: True Colors", both clocking in at about 15GB and as many hours of gameplay.
Having said that, your net connect isn't going to help with games. I have Starlink, which generally sits in the 150-250Mbit range. That should have taken about an hour to download 88GB. It took 7. The CDNs for game delivery aren't exactly speedy. (And that's pretty common, in my experience. Even in my last place when I had gigabit fiber, I generally left the console downloading overnight.)
That being said, it's a good idea to do one of two things:
1. Run some fiber to a centrally-located faceplate in each space, with MT-RJ jacks (or something else, like LC).
2. Run a conduit (an inner duct) to each office/lab space, with a pull tape, so that fiber can be easily pulled through in the future.
I personally prefer #2, for reasons like this: We have a few buildings with fiber run into offices/labs, but the fiber run was multimode. That's fine if you want to connect to a switch in the floor or building network closet, but it's not going to be compatible with the inter-building fiber, which is all singlemode.
20+ years later, laptops have 40Gb Thunderbolt and 10Gb USB, but Thunderbolt or USB to 10Gb Ethernet interfaces are still expensive, though 5Gb and 2.5Gb seem to be becoming cheaper and more common. And even a basic, unmanaged 5-port (e.g. TP-Link) 10G/5G/2.5Gb Ethernet switch is still something like $300.
Why make an app/service efficiently use 10Mbps if everyone has 1Gbps, for example?
Also, in the cloud, you're often not given sustained 10 Gbps, because e.g. the box with your VM runs 32 guests and only has a 40 Gbps link.
It is strange when some Nordic countries are already getting affordable 2.5 to 10Gbps Internet.