How much total throughput can your wi-fi router really provide?
smallnetbuilder.com
smallnetbuilder.com
I graph my data in/out with grafana. When a new IP or MAC shows up on my network I get a push notification on my phone (I wrote a small script to do this). Really great to have options to do what I want. Also it runs wireguard.
https://www.amazon.com/gp/product/B086MK4YB8
It also has a SIM slot so you can fallback to LTE if your main internet goes down.
I should mention. This doesn't have wifi. I buy distinct access points that are POE powered. For those, i use the WAX-610.
That being said, I don't know what Wifi chipset would be really good option right now.
The thing is that those kinds of speeds are not realistically possible.
Ubiquiti claims "2.4Gbps" on 5Ghz in the specs, but it's one of those marketing things, like how TP-Link claims to sell a "11Gbps" router, which is just them adding together all the absolute peak data rates together times how many simultaneous streams it can support. It's marketing nonsense, just like those wall wart-powered PC surround speaker systems that promise "3000 WATTS OF POWER".
No client device will ever get near those promised ultimate maximum speeds, because most client hardware is 2x2 streams (two receive, two transmit) or rarely 3x3 for high-end devices. The client is the limiting factor, it doesn't matter if the router is 4x4 or 8x8.
If you look at the matrix of possible data rates, 802.11ax can do up to 1.2Gbps per stream, and can obviously use two simultaneous streams, that's where they get the claimed 2.4Gbps figure. But you can only get that bandwidth if you use 160MHz wide channels, which severely cuts down on the number of available channels.
If you can detect any other 5Ghz networks from where you want to set up your AP, I would strongly advise against trying to use 160MHz wide channels, and even 80MHz wide channels can be troublesome in apartment buildings. For an enterprise deployment with multiple APs to cover entire floors and buildings, forget it. 40MHz wide channels are the realistic choice in that scenario. That's why Wifi 6E with the additional new 6GHz spectrum is much more interesting than 802.11ax.
But let's say you can run 160MHz wide channels with no interference, your client device can support it and you're close enough to the AP so your connection is strong enough to actually get that promised 2.4Gbps PHY speed. It's theoretically possible, at least.
Well, then you go ahead and subtract 40% for combined overhead in the wireless communication and protocols.
That puts us at 1.4Gbps throughput. For a single device under 100% optimal conditions.
I'm sure Ubiquiti crunched the numbers and figured it wasn't worth it to put 2.5Gbps ports on their APs, especially considering 99.9% of them will be end up plugged into corporate 1Gbps PoE switches. Can you even get reasonably priced 24/48-port 2.5Gbps PoE switches?
I've also done wireless deployments with APU2 and a wifi card, and the range was not great. Things also might be better with better antennas, or under Linux where drivers are likely more mature. I've never run anything but OpenBSD on them.
Hth :)
Hardkernel has a J4115 Celeron (10 watt max tdp) board with max 32GB ram (two sodimm slots), and 2 x 2.5Gbit ports. They also have a 4 x 2.5Gbit board you can add. So relatively low power, 6 x 2.5Gbit ports, plenty of ram for running any network services you want.
No idea how well OpenWRT or similar would be supported though.
As for 2.5Gbps Ethernet NICs with OpenWrt - I guess it depends on the particular component/NIC. People I know have had rather bad results with 2.5Gbps Realtek NICs on Windows, and I don't know if the Linux drivers fare any better. Unless you really need the additional bandwidth, I'm pretty certain that you will have the all-around better experience when sticking to 1Gbps hardware for now.
I currently have an EdgeRouter-X running openwrt and are relatively happy with it. I think I could eek more bandwidth out of it but honestly I'd rather not fiddle with it.
I have some Unifi AC-Pro access points and am pretty happy with those.
Combined router/access points devices are the networking equivalent of TV/VCR combo devices.
Basically set the trigger (the "if this") to an incoming webhook and set the action (the "then that") to be sending a notification.
https://companion.home-assistant.io/docs/notifications/notif...
It’s all software
I hugely support the idea (or just using Debian), but what a sad sad sad sad sad state of affairs there is with trying to do this!
There's the most limited, tiny, miniscule amount of wifi add-on cards one can buy with decent chipsets. Consumer offerings with consumer-targetting chips have never worked well for me (terrible performance with multiple clients, huge multi-second drop outs, low max client limits), and I spent years trying hard to make something work. The market feels basically reduced to one company, Compex, making PCIe & m.2 addons[1] one can maybe perhaps buy, that do a decent job, and their availability is somewhat scarce, and prices are pretty high.
Imo DIY ought be such a commonsense straightforward alternative to these fancy routers. Alas it seems almost impossible to do, short of scrounging for some expensive, demanding Compex chips, and hoping the open source drivers are up to snuff (they're not always). And Compex prices have gone up and up and up. Their modern 4x4 m.2 card is $400[2], but hey, at least it's possible to buy something: without this card, we'd have nothing. Oh, and that card doesn't have x86 drivers available for it, only drivers for Qualcomm platforms.
[1] https://compex.com.sg/wifi-module/
[2] https://www.arrow.com/en/products/pn02.7/compex-systems-pte-...
Think OpenWRT would work well on something like this: https://www.hardkernel.com/shop/odroid-h2plus/
It's pretty low power (TDP 10 watts or so for the CPU), pretty fast (for a router), max 32GB ram, and has an expansion board for 2.5Gbit x 4 on top of the 2.5Gbit x 2 on board.
Was hoping to hook up two desktops and a NAS with 2.5Gbit, then run GigE for cameras, Roku, etc.
I'm just using darkstat, which has the benefit of being dead simple and available via opkg. But, it's fairly limited. I already have a Grafana dashboard elsewhere on the network, so I'd be curious as to how you feed it with network stats from OpenWRT.
You should think about putting it online via github or similar. Probably many others would be interested in that functionality.
I live in a shoebox NYC apartment so a potato can give me pretty good service. The potato I picked was the TP-Link Archer A7 for $52 (at the time).
My internet speed is ~280/~280 Mbps and I have successfully attained ~440 Mbps transfers speeds between my local file server (hard-wired) and my Macbook (over wifi). So the router is not a bottleneck to the internet from a wifi device.
At any given point in time, I have ~15 devices connected to my router and it handles it great. My router supports band steering (i.e. the router and wifi device somehow jointly decide whether the device should be on the 2.4 GHz or 5 GHz band) so I get both great coverage and the best possible speeds at any given location in my apartment.
I don't see the need (for myself, your needs will vary) to upgrade to a more expensive router.
Also, does your shoebox apartment have reinforced concrete walls? That would make everything much more difficult.
That way you can get 4x4 MIMO, 802.11ac, 802.11ax/wifi-6, etc and when there's a new version you don't have to replace your router to get it. Doubly useful if any future needs might include needing a second AP.
I also strongly agree with delegating routing / wireless to different devices, having a PoE AP offers so much flexibility for positioning so you get better coverage.
I just moved into a larger house and after some research decided to skip the $99 Ubiquiti 2x2 wifi6/802.11ax and get the $179 4x4 Ubiquiti wifi6/802.11ax, just received it this week and haven't installed it yet.
PoE is also REALLY useful for cameras, hoping my 8 port PoE switch does well with 2 APs, and 5 cameras.
It's not required though. I just run their AP management software locally through docker. Said management does not even have to be running all the time, the APs will happily run the config you push to them without management.
So no internet access required for their APs. The GPL violations does worry me, hopefully they get better on that front.
https://www.engadget.com/2016-08-21-nsa-technique-for-cisco-...
Is there any evidence behind this assertion or is it just the usual fearmongering and baseless speculation?
On another topic, pretty devestating that OpenWRT runs on only one of these routers, the slowest/oldest one, the Netgear Nighthawk X4S aka the R7800, which dates back to 2016.
In that they need technical specifications and currently-closed source firmware, in a low-margin business, where providing those is last on most companies' priorities?
Or at least, those are needed to enable the latest-and-greatest in a direct hardware path.
Ultimately lot of the issues fall down on Broadcom and Qualcomm being what they are and not even trying to upstream their drivers and other bits.
I don't know if we ever will see another wifi chipset that will have as good and long support as ath9k.
The original WRT54G, for which OpenWRT is named, shipped with a Broadcom brcm47xx chip[1]. But something changed in the last decade, and Broadcom has been unsupportable, unusable, & there is no community, no folks putting their own Linux onto Broadcom based systems for a long time now. I would love to know the specific details, to know what happened. But the status quo is that Broadcom is the worst villian, is unsupportable, that nothing they do is generally usable by hackers & Linux users, and that any router with a Broadcom cpu in it is utterly unusable.
It's worth mentioning that the Broadcom wifi drivers on Linux have been a constant tussle & difficult. For the longest time the MacBook's worst area of support when running Linux was the wifi drivers, owing to Broadcom's dis-support.
I hope some day Broadcom stops being such a dark side impossible company. But there've been no points of hope that I've seen in the last decade. It's just gotten worse. Maybe it's us, maybe we haven't tried hard enough to see what access we might be able to get, maybe Qualcomm has become the main OpenWRT, Debian, &c running chip because it's what folks knew worked, because the long history of Atheros (who Qualcomm bought) being moral & good folks, who understood the relevance & importance of support, and we just haven't tried hard enough to force access to Broadcom, and that's on us. But right now, the status quo is, Broadcom chips are all useless to the world, aside from whatever software comes on them, and whatever software & support you get out of the box. A lot of people are very sad about this, and I recommend also being sad about this.
I was much happier with the Ubiquiti stuff, ipv6 just worked, regular easy updates, and it's been solid.
your mileage may vary. personally i'd rather be running Debian, and have never quite gotten the spare time to try getting Debian going on these systems (i would borrow the kernel & maybe some other packages from openwrt, maybe even kexec in to Debian from openwrt). but openwrt has been low muss low fuss for me since I installed it on a WRT54GS a decade & a half ago. there's certainly some initial uncomfortability learning uboot, understanding what to do. but i've never had to go to any special measures for things to be just working.
did you report any of your issues? are there bug reports for anything you encountered? nothing in what you wrote up sounds right or good or like what i've experienced or would want to. i hope some of these problems got captured.
Ah, yeah, mine was the ac1900 chipset, the r7800 is the newer ac2600.
Lack of Funding, Ads revenue and Interest. The owner has mentioned this multiple times.
The author visibly knows their stuff and goes into channels, ac, ax, 2x2, 4x4, MU-MIMO, beamforming, etc in great detail.
But the summary of what I took from that is the times when consumer routers were bad are mostly over.
Modern high performance Wi-Fi has a lot of radio magic, so your two-antenna laptop can talk to two antennas of router via two multipath-propagated spatial streams while the router uses its two more antennas to form directed beams and get more signal so more range and speed.
All of that needs a lot of signal processing magic, and is heavily proprietary and patented, so at the moment it won't work well on non-blob firmware and drivers. And even small manufacturers have trouble here - in pure Wi-Fi performance, Mikrotik would be behind Asus or Netgear or something, even if it has more features and more professional interface.
Of course, you only need to care about that if you need local-network performance or you internet connection speed approaches Wi-Fi speeds, otherwise it doesn't really matter.
Fixating on throughput benchmarks is how you get bufferbloat.
Assuming that the path under test latency is small to begin with, increasing buffering from reasonable to large to absurd generally provides a tiny and rapidly diminishing throughput increase.
Part of the reason bufferbloat exists is because of norms in equipment testing focusing on throughput (esp for single/few tcp streams) and drops while ignoring other effects.
Installed dozen over years and can say only positive things about. Also, very flexible with configuration.
One headache less.
> Whistleblower: Ubiquiti Breach “Catastrophic”
> Trust me, this whistle-blower "Adam" (I have a few suspicions of who it actually is), toned it down. > The reality is much much worse.
I'll be adding one or two APs and managing them with CAPsMAN when we move to a larger house. IMO it beats Ubiquiti's offerings on functionality, but is behind on polish.
I just went through a battle helping a friend who is a streamer & who lives in an urban setting with hundreds of ISP provided 2.4 and 5GHz APs visible from her apartment. She had horrible glitching issues when live-streaming which I guessed were due to latency spikes when other APs were talking on the same frequency.
I tried to research what routers support DFS, and settled on a Netgear AC3200 which they claimed supported DFS. However, we could never get it to select a DFS channel, and there was no mention of DFS in the firmware. After a week of useless Netgear "support", I just gave her an old TP-Link 5Ghz router I had laying around. It was running dd-wrt (or openwrt, I can't recall), and had no problem selecting a DFS channel.
Since then, her streams have been nearly perfect.
I just wish DFS was a feature these reviews tested
Businesses are more interested in reliability, security, price, details of the warranty, support contracts, and other details that don't show up in benchmarks.
If performance is key, they buy samples and test in their workload. Or at the enterprise scale, they might get loaner units from their vendors to evaluate. Benchmarks rarely tell the full story for unique workloads.
It's possible to approximate the https://dslreports.com/speedtest using the flent CLI or QT GUI (which calls e.g. fping and netperf) and isolate out ISP variance by running a netperf server on a decent router and/or a workstation with a sufficient NIC (at least 1Gbps). https://flent.org/tests.html
`dslreports_8dn`: https://github.com/tohojo/flent/blob/master/flent/tests/dslr...
From https://flent.org/ :
> RRUL: Create the standard graphic image used by the Bufferbloat project to show the down/upload speeds plus latency in three separate charts:
> `flent rrul -p all_scaled -l 60 -H address-of-netserver -t text-to-be-included-in-plot -o filename.png`
In 2021, most routers - even with OpenWRT and hardware-offloading - cannot actually push 1 Gigabit over wired Ethernet, though the port spec does say 1000 Mbps.
So you could build yourself a router, run OpenWRT, pfsense, or whatever floats your board to control which packets go where, and which ports are visible from where. The APs do not need access to the internet, and generally anything sensitive going over the network should be encrypted anyways.
I'd consider a router I control the OS and configuration of with a ubiquiti AP (or 3) more secure than the usual router configured by some corporation or running a bunch of wifi binary blobs.
You mean shit locked into a proprietary cloud service that delivers you new vulnerabilities at arbitrary times?