I heard Atheros is a lot better than MediaTek and etc. Any recommendation for the cheapest in the market?
You will not have fun with the cheapest device if this device does not meet your requirements.
What are your requirements?
Why are you searching for the cheapest?
What volumes are you talking about? 1, 10, 1000 pieces?
Interesting. I'd look into that further before making a significant buying decision (I don't consider a single US $10-15 router in that category). I've been wondering if it's not the other way around now, with MediaTek having been pushing pretty aggressively at a given price point now for several years.
on tmomas' statement -- decide your requirements, then determine the most effective solution. Saving $0.50 to only spend dozens of hours trying to shoehorn what you need into the device, only to find out that it crashes every hour since it doesn't have enough RAM is, to me, false economy.
Unfortunately, most "amateurs" do it the way, the opener of the thread does it.
Actually, I'm very sure about what I want, I am building a public WiFi network with min cost for anyone to participate (low cost of ownership). Thus, I was told Atheros chipset produces the most stable WiFi routers, hence I'm asking the question. If WiFi routers can be so certain of their performances - number of devices supported at what speed, what dBm, etc, then maybe my question would be more precise.
Yes, I'm an amateur, but an amateur has to start asking somewhere somehow. You'd do the same if you know zero about how to build a startup. However, I'd never label you as an "amateur", instead I'd only be encouraged by you having the guts to ask "stupid" questions.
Or maybe I should rephrase. From my 3 month of testing OpenWRT on 4 different devices, I noticed that OpenWRT could never equal the performance of the respective stock firmware. So, Jeff and reinerotto, how could I be certain of my requirements, when OpenWrt couldn't be certain of its performance off the shelf. I'm not talking about having to mod the image.
Some specifics as starters:
- Do you need 802.11ac?
- Do you need IBSS and/or 802.11s support?
- Do these devices need Ethernet ports and, if so, how many and do you need GigE or will 100 Mbps be sufficient?
- Is a single phy sufficient, or do you need two (or more)?
- Are there any CPU-intensive tasks required, such as bandwidth shaping or VPN?
- Is there a physical size limitation or other form-factor consideration?
- Are external antennas (ones that stick up/out) OK or not?
- What power requirements do you have (wall wart vs. microUSB vs. integral mains plug vs. PoE)
On "OpenWRT could never equal the performance of the respective stock firmware", what aspect of performance did you find lacking in comparison?
Let's talk about an average WiFi router that has the most common specs that is still in production.
Well, since you listed out...
- Do you need 802.11ac? -No
- Do you need IBSS and/or 802.11s support? -No
- Do these devices need Ethernet ports and, if so, how many and do you need GigE or will 100 Mbps be sufficient? -Min one 100Mbps port
- Is a single phy sufficient, or do you need two (or more)? -Yes
- Are there any CPU-intensive tasks required, such as bandwidth shaping or VPN? -No
- Is there a physical size limitation or other form-factor consideration? -No
- Are external antennas (ones that stick up/out) OK or not? -Ok
- What power requirements do you have (wall wart vs. microUSB vs. integral mains plug vs. PoE) - Plug
I found that most could not handle as many concurrent devices as the stock.
It sounds like most of the "cheap" routers in the post that tmomas linked would fulfill those requirements, assuming that they're from a manufacturer that is likely to be in business for a while and have at least 16 MB flash and 128 MB RAM. (I'd stay away from low-flash or low-RAM devices for stability and to allow for future upgrades.) I know that there is at least one for under US $20 that appears to fit those needs (though it is powered over microUSB).
When it comes to range, RF components are relatively expensive parts of a device's BOM, and things like power amps and low-noise front ends often get dropped early in the price-reduction process.
Personally, I've never had an issue with number of concurrent devices. Perhaps you can explain more of what you're seeing.
On MediaTek vs. Qualcomm, while the early open-source MediaTek wireless drivers were dicey, from what I've read there has been a great deal of progress over the last year or so and the current drivers are good. Many thanks to the devs on that. A search of the forums here will provide background and current status.
Agreed. I developed and put several different open Wifi networks into production, most of them based on MT7620A. Work like a charm. Even with lot of additional software, like captive portal, php, nginx, squid, wireguard, openVPN ...
What devices you actually used? What models?
How many concurrent devices you tested? 5 or 15? How you tested? Streaming? Random surfing via bots? What routers you tested?
I tested on Xiaomi Nano, TP-Link WR741N, 841N all with the latest hardware versions that are supported by OpenWrt.
7 devices - phones, laptop, iPad, smart TV
Testing was done by concurrently streaming from YouTube, Twitch, DailyMotion or other streaming sites, web browsing.
Internet plan is 100Mbps
Most routers couldn't handle more than 5 devices. Errors such as inability to connect to WiFi and basically all devices couldn't go online until I have to reboot the router. Xiaomi performed the best, up to 7 devices, but occasionally the WiFi just stopped working and all streaming and browsing stopped until reboot.
NOTE: Now, when I tested on the stock firmware, I dont have such issues.
OK, let me start by saying I'm hardly an OpenWrt fanboi. However, I believe in using the right tool for the right job, and OpenWrt is probably the best tool out there for consumer-grade wireless routers when you need security and stability.
My main router is a Lanner FW-7582 that has been upgraded to an Intel Xeon E3-1265L, because there isn't a consumer-grade "router" out there that can come close to handling the load here. It runs FreeBSD 11.2. Main pathways are all bonded GigE and switches are all Cisco. My file server runs FreeBSD with ZFS with both L2ARC and ZIL caching on Samsung/Optane SSDs. It can easily swamp bonded Ethernet connections over NFSv4 mounts.
I run five Archer C7 v2 units as APs, each running multiple APs, two 802.11s meshes, and wireless backhaul of multiple VLANs. I regularly have 20-40 devices active on my network, most of which are concentrated on two of the C7s. (The others provide remote coverage outside of the house.) We often push over 200 Mbps, sustained, over wireless to/from individual clients, which means 400 Mbps over the air.
Can OpenWrt reliably handle that on decent (750 MHz, single-core MIPS, dual Ethernet phy with 128 MB RAM) hardware? Absolutely, and at probably 5 times the throughput you're running through your 100 Mbps line and 100BASE-T Ethernet interfaces can put through your devices.
Here you go, run just now, 12 concurrent streams, 300/30 is what my Comcast line is good for
So what's "wrong" with what you're seeing?
Well, a year ago I would have confidently said, "No open-source software is ever going to perform as well as OEM software in a typical home situation." Why? Because for years the manufacturers have off-loaded the NAT to the switch chip using proprietary drivers, bypassing the Ethernet phys and SoC completely. Just servicing the interrupts at 300-500 Mbps throughput can bring a typical MIPS processor to its knees.
Now, however, there have been open-source drivers for the NAT functionality in many switch chips (see "flow offloading" threads here) that have, as I understand it, been mainlined for many devices on the 4.14 kernel. That gap should be gone, assuming you can live without bandwidth flow-shaping ("SQM"). But hey, none of the OEM firmware can do that anyways. Even for some of those that claim bufferbloat reduction or the like, read the fine print, you need to install special drivers on your clients.
Now, the next problem -- insufficient memory.
Let's look at those devices you selected, and how they stacked up to
- Xiaomi MiWiNano -- 64 MB RAM, 575 MHz MIPS
- TP-Link WR741N -- 4/32 device, 400 MHz MIPS, apparently you didn't read https://openwrt.org/supported_devices/432_warning -- it's an obsolete piece of crap
- TP-Link WR841N -- 4/32 device, 560 MHz MIPS, slightly better, but still another one to donate to charity for the $2 tax write-off.
No wonder they hang under load.
Think I'm off on that? "Xiaomi performed the best" -- the only one with enough memory to even consider.
Not to mention that they're all crippled by 100BASE-T Ethernet.
Hey, but have at it, run OEM software with an unpatched 2.x kernel and skeletal application software, all with known vulnerabilities that are likely being exploited in the wild. Just keep it off the Internet so you don't do others harm when it's compromised and used as a jump point or C3 node.
You're missing my point here. Like I said, I'm comparing OpenWrt with stock firmware. How can my router be a "crap" when it performs flawlessly with it stock firmware.
You get it?
I can concurrently run 15+ devices at once with the "crappy" routers you said with its STOCK FIRMWARE. Don't you get my point? I'm comparing speed and realibility performances of OpenWrt vs stock firmware. Dont go and change the subject on vulnerabilities. We can take that on a different topic.
OpenWrt might as well take out those "crappy" routers you guys said off the list if you think they work so crappily under OpenWrt. Thus I don't have to bother wasting so much time/money testing so many routers.
You were given an opinion about the device with regards to OpenWrt. What does OEM have to do with it.
I'm giving my opinion in return on their opinions based on the comparison between OpenWrt and OEM. I'm not for or against OpenWrt. I just need a solution and I need to know if OpenWrt could perform as expected (benchmark is stock firmware) on low-cost devices.
how many clients were possible with stock firmware?