The coolest hardware in the cheap router/APs that run OpenWRT is the programmable Ethernet switch. VLANs on such cheap hardware is an awesome tool to split up stuff at the ethernet level without having to have a bunch of NICs. This doesn't seem to have it. Does anyone know of a 5 or 10 port cheap switch that also has programmable VLANs? Those would be great for homelab kinds of uses.
If you want a serious switch, you can often find a big name used switch for $100 or even less. eBay has a couple of 12-port Ruckus switches listed right now, for example.
Oddly, if you want to purchase a license to enable the fancier capabilities of Ruckus switches (those “12 port” 1Gbps switches actually have two 10Gbps ports), it’s surprisingly awkward to find anyone selling the licenses to individual users. On the other hand, the documentation quite strongly suggests that there is no verification whatsoever if you simply tell the switch that you have the appropriate license.
(I assume that there are a few things going on here. First, big users who buy tens or hundreds of these switches will have accounts and will actually pay for licenses as needed and presumably get decent pricing. Small users are basically irrelevant to the manufacturer’s bottom line. And serious users will regularly use in a configuration in which they don’t have Internet access, and will quite sensibly refuse to purchase the switch if it needs Internet access.)
(Whoops, the campus network is down because the core switch needs to re-validate its license. But it can’t re-validate its license because it can’t access the Internet until after it starts working again. Better choose a new vendor next time!)
But do they run FOSS software? And can they thus be trusted? Lots of network gear gets zero day'd and exploited. I can understand the sentiment of wanting to run something that can potentially be fixed and fixed quickly.
Running FOSS software on a switch is an awkward endeavor. Some switches have okay-to-decent support, mostly via OpenWRT, and you’re mostly getting VLAN control. With Ruckus or Cisco, etc, you also get ACLs, some “layer 3” capabilities, sFlow, SNMP, real support for various loop detection schemes, network mapping capabilities, possible diagnosis of cable problems, and lots more. And a configuration system that is quite a bit better than you will find in most Linux networking config software. (Although OpenWRT actually tries pretty well, in contrast to, say, Ubuntu. Sigh.)
I would not want to run a large network using OpenWRT switches. Maybe if OpenWRT took management of multiple devices seriously some day.
FWIW, in theory you can run a mostly-open Linux stack on some of the very software-defined switches, supporting OpenFlow and such. See, for example, https://github.com/sonic-net/SONiC/wiki This did not seem like an easy thing to get working.
I really wanted to get whitebox switches for my lab. There is very limited switch model support out there and almost none of the supported models are suitable for home use. No one needs a $6000 100gbe switch for their home lab. I need an 8 port 10gbe switch or a 48 port 1gbe poe switch and you will not find support for those types of switches from the white box operating systems. Whitebox switching is for datacenter and hyperscaler sdn. It’s not like building a whitebox nas or rolling your own bsd router.
There’s really only two managed switch options for the homelab in my opinion. Mikrotik or, if you don’t mind loud noise and higher power consumption, used enterprise gear e.g. cisco.
I’ve migrated all my network stuff to vyos routers on used sophos boxes, and mikrotik switches. The switches do nothing besides vlans and mc-lag. Vyos is great and supports everything I need like wan load balancing, vrrp, bgp ecmp, dhcp, dns forwarding.
FWIW, at least some of the small- and medium-size Ruckus switches are fanless or can be set to run in a mostly fanless mode. Doing the latter may limit the available PoE output.
Also, eBay is coming up with plenty of reasonably priced Dell -ON switches and some very very cheap EdgeCore gear, but they might be old models and don’t appear on the Pica8 hardware compatibility list. Sigh.
I stand corrected, looks like you can get some of those dell switches for a couple hundred bucks, haven't looked at the market in a couple years. To me the biggest thing mikrotik offers that no one else seems to is small reasonably priced 10gbe switches. I don't need a 48 port sfp+ switch, I need two 12 port ones with mc-lag for my ceph/k8s cluster.
This has been one of the most rewarding comment threads on HN I've read in a long while. Lots of learning even for me in terms of what the current state of things is in terms of cost and choice.
I've used a couple of these $30 8 port managed switches https://www.amazon.com/gp/product/B00K4DS5KU/ another commenter mentioned for various things and been pleased. This is fanless, cloudless, and has web-interface configurable VLANs, LAGs, mirroring, jumbo frames, IGMP, and STP.
You can use a managed switch to get this working if you have a router with too few ports.
I have several gl.inet routers running openwrt but they all have 2/3 ports total. I put a TL-SG108E in front of them and use that to tag ports.
This particular switch has an interface that is quite easy to reverse engineer, so I have written a script that allows me to easily move ports between VLANs without bothering with the unintuitive web UI.
The Zyxel GS1900 series switches run OpenWrt well and like can do VLANs. You can an, 8, 10, or 16-port one for less than $100 on eBAY. The GS1900-10HP has 8 GigE ports with POE and 2 SFP ports and I'm seeing it on eBay for around $60 right now.
I am really excited about this one. I have used the Turris Omnia in different settings, and can absolutely recommend it for people that want a great supported Open Source Hardware and OpenWRT router. This one seems to be an alternative to the Turris Omnia.
Isn't the omnia like $350? For me that's prohibitively expensive to spend on a router, whereas the projected $100 price point here is much more palatable (I probably still won't be buying it because I need SFP, but if not I would consider it)
So, what are your options?
I'm in the market for a "cheap enough" router supported by OpennWRT with an SFP because my FTTH is terminated in the ISP router currently (but that router sucks, even if I moved the wifi network completely outside it, it still hangs from time to time)
I’m currently using a Banana-PI R3 which meets those requirements. It only recently got release support for OpenWRT, but SinoVOIP seems to be actively working with them (which is confirmed with this announcement). There are still small issues (last I checked the boot loader method of cloning to nand is broken), but it’s been pretty solid for the last 5 or 6 months.
I was coming from a late model AirPort Extreme but needed something with more configurability. My biggest with it issue is that the BPI is a fast computer with lots of RAM and storage (and SD and M2 expand ability), but OpenWRT is made for the tiniest embedded systems and everything from configuration to pre-built packages prioritize space over quality of life. Busybox is fine, but when I have as much power as a desktop computer from 10 years ago, it feels like an unnecessary inconvenience. Or as I just went through, Ruby installs the Ruby interpreter and nothing else, each module in the standard library is a separate package and someone went through the work of configuring the dependencies of each of them. Likewise, LUCI is great, but sometimes I just want to configure a thing, not have to find the LUCI configuration for the thing.
This reminds me of a recurring thought I have had over the past 15 years using UNIX-like OS. In the past, there were many more computer architectures commonly in use, and some UNIX-like OS projects made efforts to port their OS to a variety of architectures. (Not naming names, but please note I am not referring to Linux.)
Today there are fewer archs in common use, but there is still variety in hardware. UNIX-like OS projects make efforts to port their OS to a variety of hardware. Assuming they make an attempt and are successful, this usually takes time. There will usually be a considerable period of time where the hardware is for sale but before the non-Linux UNIX-like OS project is running on the hardware, and able to utilise enough of its features to justify installing it.
And so the recurring idea I kept having was "Why not prioritise support for a particular hardware product."^1 Of course this idea is unpleasantly inequitable, but the practical benefits could justify it. As it happens, eventually, the project I was using started to officially prioritise certain architectures.
1. The hardware product chosen for prioritisation should be one that is likely to be produced for many years, not the type that is a passing fad.
The benefit I imagined is that someone who prefers to install a non-Linux UNIX-like OS themselves could purchase a new item of hardware and install the OS on it and have every feature working _immediately_. Generally, this would be potentially 100% of people using this non-Linux UNIX-like OS. There is no GUI, all are capable of installing the OS with or without an installer, all understand how to compile software from source.
To some extent, this already happens. There are some particular hardware products, e.g., development boards, that are produced for many years that continue to work with this non-Linux UNIX-like OS and they tend to receive consistent attention by the OS project volunteers.
To sum it up, the pipe dream here is that some non-Linux UNIX-like OS user who does not like pre-installed OS and likes to install the OS himself can purchase a brand new device and _know_ 100% that all the features of the hardware will work.
I have had good results with the GL.iNet routers too, but the "supported by openwrt" just means that GL.iNet pulled an old version of OpenWRT and hacked it up until it worked on their board instead of making their own OEM firmware. They're not upstreaming their patches or collaborating with OpenWRT, just kind of leeching off them.
I'm not even sure you can install vanilla OpenWRT on the Flint 2 yet.
In contrast, this is made by and with the OpenWRT dev team!
It looks like the patch was committed in late September [0] and the press release for the device is dated mid November [1]. So unless I'm reading things wrong they added support upstream before they released the product?
Was looking at this recently, but couldn't find any sort of range information. I have the Beryl (wifi 5 version) and it's good for a hotel sweet and that's about it. I would hope the Flint 2 would be better, but I can't find anything.
"The router’s specifications have been selected with the goal of keeping the price under $100, and that’s why we have interfaces such as USB 2.0 instead of USB 3.0 since there aren’t any spare ones in the Filogic 820 SoC"
2.5G is still plenty for most creator use cases (even for editing a few 4K livestreams off a NAS). The only difficulty is if you hit certain types of routing, performance can't always keep up (sometimes even to a gigabit).
One thing I'd love is if more vendors would give a bunch of examples and specifically list what things are hardware accelerated and which are not (MikroTik's been pretty good about this lately, though it's a lot of digging in docs).
What is the difference between a "homelab" and a "creator"? I don't think these labels mean anything.
Also you realize this is a router to connect to the internet right? If you want computers to have faster ethernet links between themselves you can have a switch with faster ports. It isn't going to make sense to try to mix your router and your switch at that point.
I really want to like this, but what interests and concerns me more is the finished product, specifically WiFi range. I have a TP-Link AC4000 which doesn't support OpenWRT, but I can have like 20 devices connected throughout my 4 bedroom house and 30m from the AP I can get about 500Mbps throughput at 0% loss, and the throughput is probably more a limitation of my ISP. Not sure I'll find anything better than this at the price point I paid ($150AUD).
What are people doing that their homes need so much bandwidth? My ISP offers Gigabit and I use their cheaper, third-tier, 200 Mbps option, since I cannot conceive of a use for more.
Off-site backups is a huge one. General downloads for the people running *arr software, and corresponding upload for people sharing their Plex libraries.
Here in france, I get 5Gb/700Mb for 50€ and there isn't a lower tier at the same ISP. Other ISPs would simply sell me 1Gb/700Mb for the same price, with no lower tier other than aDSL. Not that I'm complaining, but if I'm getting 5Gb, buying a router for 150€ that hits barely 1/5th of that speed (the maximum on wired this board can do is 1Gb, and I don't even have all my devices on 5GHz wifi) isn't exactly a good experience.
Here in the UK I'm still stuck with BT who keep sending out flyers and public consultations saying they'll upgrade for about 5 years now, I'm assuming to try and dissuade altnets from coming through and taking away all their customers. They keep silently killing the plan after a year, and replacing with a new one for 2 years time so it never arrives.
50/15, not even that far out of a major town. Fucking BT.
One of those cases where being first is a disadvantage. We've got cable, and it's been upgraded some, but because of that there's like zero incentive to get fiber.
Meanwhile my buddy is out in the bushes and he got 1G fiber like you a couple of years ago...
The ports aren't backed by switching hardware, so would have to get bridged in hardware. If you don't want to use it as a router towards the internet, just get a 2.5G switch would likely be more performant and future proof given this only has two RJ45 ports.
Like the parent, I'm annoyed at just Wifi 6, when 6E is out and 7 is around the corner. However, the feature of 6e (and 7) that I need is the new, larger 6Ghz band for stable performance in a crowded urban environment with hundreds of ISP configured routers all crowding the same few groups of channels on the 5Ghz and 2.4 Ghz bands. For me, 6ghz is far more stable than 5Ghz (or 2.4).
Really excited for the news, but then really dissapointed at the same time.
I thought it has several RJ45 and SFP ports, similar specs to Banana Pi R4 based on the provided picture, apparently it is only for illustration purpose [1][2].
Hopefully this will spurs more complete support and drivers for the BPi boards including the R4.
If this one does well, I would expect further developments. Bananapi has the capacity to do lots of board designs, they have a huge backlist; so it's down to the openWRT team to see how many they have the capacity to do with them and support.
This could be a good match, because bananapi do a lot of boards but their software story has been a bit poor, as it tends to be an android build hacked together with a conventional linux distribution (or was a few years ago when I last looked). Whereas openWRT know how to build a software platform.
bananapi have some kind of link with foxconn, but I don't know what kind.
> bananapi do a lot of boards but their software story has been a bit poor
This is quite common with other board manufacturers too. I'd rather suggest to ignore completely their cobbled together distros, often also tainted by proprietary modifications, that become unmaintained in a few years, and see if they're among the many supported by Armbian or DietPi.
I had a good experience with Armbian, it was my first time modifying a linux distro with extra packages so that we didn't have to run install scripts and compile on the orange pi (actually I started looking at how to build armbian because the build step for python's crypto module would fail after cranking away for an hour. Not enough memory or something timed out. Anyway running the armbian build process with a target cpu architecture was a breeze following their docs)
SINOVOIP is the major name associated with BPI, but it does seem like it is also affiliated with Foxconn. At least one of their engineers has been working with the OpenWRT team closely on the R3 and has been a wealth of knowledge in the BPI forums.
A doubt - isn't the antennae an important part of any Wifi router? Are they too standardised now, with multiple options, and as easy to use as plug-and-play with such boards?
There are basically three types of tiny push-on coaxial connectors in common use. You can get antennas for any combination of bands with any one of the connectors. You just need to be careful to order the right combination.
> The Banana Pi BPI-R4 can be purchased for $103.15 plus shipping on Aliexpress, but crazily enough this does not include the WiFi 7 card which I can’t find for sale anywhere yet…
Not the same price, since it doesn't include the card.
But I for one wish the OpenWrt One had a higher price in exchange for Wi-Fi 7.
I'm not the person you replied too, but I could see myself using it to run object detection from my security cams. Frigate NVR specifically recommends a small ARM board with a Coral TPU.
How paranoid should I be about routers from companies like Cudy that I’m not familiar with? My intuition is “very paranoid, even if I replace the firmware with OpenWRT”.
Yes, you are being too paranoid. Once booted into Openwrt the router will only be running Openwrt code. I guess there's the tiny possibility of a backdoored bootloader but that would have to be a pretty sophisticated backdoor!
What are those 3 slots on the top left? Looks like 3 slots for sim cards, but there's no mention of SIM support and only 3 quantities of something are antennas.
That looks like it's the Banana Pi BPI-R4 WiFi 7 router board: https://www.cnx-software.com/2023/11/23/banana-pi-bpi-r4-wif.... And yeah, those are three nano SIM card slots. You can see the image of the underside with a bunch of places to attach the relevant antennas.
If there is any way you can run a few ethernet cables as a backbone, I think it's pretty simple to just setup additional OpenWRT access points on the same SSID to allow client roaming.
Or, with dual-radio routers, setup WDS between the routers using one radio per router and clients on the other, so traffic does not compete for the same channel. I am not sure whether you can still do roaming this way as I've never tried it. But I did use it in the past to brdige wired clients from a secondary router back to the main router over WDS.
Wood is not either. The inside walls here are 100mm wood(CLT) with gypsum on both sides and the signal drop through 1 wall is big. 2 walls almost kill the signal. I don't have any wifi reception outside the house at all (65m² is floor area) using 1 UI Wifi 6 LR on each floor (with the U6Pro it was even worse).
The chip would need to be powerful enough to process packets to do advanced QoS (like SQM) at line-rate. However, offload is supported on MT chips, so you can still perform hardware NAT functions, just not as configurable as those other methods for traffic control.
Maybe some VPP integration into the proper distribution, instead of packages and custom compiles, would make that work.
With the extremely important difference that it has a dual core A53 instead of a quad core A73. I'm sure the board you linked could handle QoS just fine.
There are other differences, too, including RAM and interface line-rates.
It was lazy of me to reference the other system without a deep-dive, sorry. I was trying to answer the question regarding QoS as different from offload, where the CPU might be the bottleneck in the former.
> I'm sure the board you linked could handle QoS just fine.
A quick Google shows some complaints about line-rate on the BPI-R4, too, even for a simple traffic test.
