Future of 432 devices with well developed drivers against new unsupported 5GHz devices

FYI: Strong seems to be a rebrand mark of MTC hardware. In USA exist HooToo and should be exactly the same. Would be great if someone could verify it.

Salu2

I have one of those gl-inet devices. I haven't pushed it hard, but I'd be surprised if it couldn't do the full 100Mbps routing without any flow offload (now SQM is a different story)

There are 14 versions of the TL-WR841nd (https://openwrt.org/toh/tp-link/tl-wr841nd) so let's assume you're talking about one of the older ones, like the v7 or something, it's a 400MHz AR7241-AH1A chip from 2010 or so. Compare this to the gl-inet a QCA9531 at 650MHz. So if you are able to push nearly 100Mbps on the old device, undoubtedly you're going to do it even better on the GL-iNet

1 Like
  • But you won't get a lot of throughput depending on CPU
  • If you disable WiFi or live in a rural area, entropy may be too slow to bring up WG in a reasonable time
  • Also, my point was only regarding flash space

@lleachii: I have more than 25 MBit/s with this old devices and can't test higher bitrates because of my upload into the internet is not faster. - So even if this is the limit, I can't reach more with newer devices and I think most people are satisfied with 25 MBit/s upload and have also no faster Internet-Connection :wink:

1 Like

In my opinion all the fancy hardware offload features are bad work-arounds for operating a router with too weak a CPU for the intended use-case. Now at 10Gbps and above we might have no alternative, but even a lovely wndr3700v2 will happily do traffic shaping at around the request 70-80 Mbps rates, something currently incompatible with hardware offloads AFAICT. IMHO operating 4/32 devices should be restricted to use-case where they are not out of their league, like dumb AP or similar. This will put decent and stable wifi drivers/hardware to good use and avoid most of the issues from overtaxing the limited resources of these devices.

4 Likes

Understanding just what OpenWrt is may help to alleviate the panic around 4/32 devices and explain why they are no longer viable for "general use".

OpenWrt is a Linux distro, not a complete, from-scratch firmware.

Like any other distro, you start with the Linux kernel sources and upstream application software sources. They are, within reason, what they are and are and aren't drastically modified.

OpenWrt makes some high-level choices related to balancing image size and performance against resources in its selection of packages to use. For example, busybox, dropbear, and dnsmasq are "lighter" than their full-featured counterparts that you'd find in something like Debian.

In some cases, OpenWrt-maintained or OpenWrt-supported alternatives are used for core OS funcrtions. For example, procd and ubus are sufficient to meet the needs of most all-in-one router users and are much "lighter" than, for example, systemd and dbus.

There are also "generic" functions, such as the binaries from hostap that are required to get the functionality from wireless devices that users of all-in-one routers expect.

These are "fixed costs" that are common to every install.

Once down to a platform -- a specific SoC, for example, the kernel needs to be configured to be able to boot and run generic devices, like Ethernet, switches, wireless chips themselves, LEDs, switches, ... At the platform level, basic decisions about the kernel configuration are made and the kernel configuration, at least as far as I understand, is common to all boards on the platform. Discontinuing support for a specific 4/32 board that uses, for example, a common SoC, doesn't impact this at all. If, for example, a QCA9331 is supported by the ath79 platform, it doesn't matter if it is a 4/32 board or a 16/128 board.

At the board level, someone needs to define how the components are "wired up". With the exception of x86 architecture, there usually isn't a general way for the kernel to discover this on its own. For example, "If you twiddle GPIO 20, the red LED goes on and off", or "There's a generic SPI NOR flash on GPIOs 17, 18, 34, and 45." Older kernels defined this in C, current kernels use a "Device Tree" to define it. Once defined, it's good until the manufacturer makes a change in the board. Here, if the DTS (Device Tree Source) exists, discontinuing ongoing support for 4/32 devices won't remove this and you'll still be able to build OpenWrt for the board.

So, what if it doesn't fit into flash?

You need to decide what you can remove. The "tiny" target variant makes some generally acceptable changes for most users to trim kernel size, such as, as I understand it removing debug symbols.

Now what?

For me, go ahead and kill everything related to PPPoE and the like. My modem supplies direct connectivity. By my choice won't work for anyone that has an ISP that uses PPPoE. You may only have IPv4 connectivity, but what happens if you strip IPv6 for a user that only has IPv6-based transport? I could care less about LuCI, but many users find LuCI a huge value to them.

These are personal choices, and it's impossible to have a "one size fits all" kernel / set of packages on flash-constrained devices that can get a device up and running for all users. It's also unreasonable for the OpenWrt project to create ten or twenty variants for each use case of these under-resourced devices.

What happens when the device crashes because it only has 32 MB of RAM?

Again, you need to strip out functionality, a personal decision.

Removing explicit support of 4/32 devices doesn't mean that they can't be configured and built by individuals or organizations that want to use these existing devices in ways that acknowledge their hardware limitations. Yes, if you've got a new 4/32 device, well, you may have to port it yourself and shouldn't expect that the port be accepted by the OpenWrt project. Then again, as @suppenkasper0815 notes, the cost of hardware components are so low that you've bought a "race to the bottom" device, rather than a better-informed choice at a similar price. This is a messaging issue that is being addressed.

Edit: This is nothing new. I go back to the original WRT54G and the release of the variant that dropped to 2 MB of flash and 4 MB of RAM only ran VX works. The issue of 4/32 devices becoming unsupportable has been in "public" discussion here for two years. "You can't say you haven't been forewarned."

5 Likes

In my experience with QCA SoCs, any decent, 750-MHz class, single-core, MIPS-based all-in-one router can handle 100 Mbps, usually including SQM of moderate complexity, such as "piece of cake". I would imagine that the comparable MTK SoCs have similar performance, but don't have any personal experience with them.

In both cases, this is assuming you haven't loaded the router down with all kinds of non-router functionality. This is primarily due to the limited processing power of MIPS-based SoCs at a given clock speed and will limit the device no matter if 4/32 or 16/128. If anything, a 4/32 device is more likely to have a 350-MHz class SoC, which will struggle with 100 Mbps.

1 Like

Not supported anymore, thats mean security vulnerabilities. 3.16 will follow next year. This is the best way to build a botnet...

Im using around 15 different devices and they are stable. Only serious things happening for me is the hardware dying here and there and its surely not OpenWrt's fault. Bugs happens (like in any software), you cant blame folks working for free for you.

Say thank you to the manufacturer doing a terrible job. The linux community is just trying to fix a mess here.

Sometime stuff need to be patched after software beeing roled out on a massive scale. A piece of software is NEVER truly finished and bug free. Some releases has been rolled out because of big security vulnerabilities.

2 Likes

This is further complicated by trying to support hundreds of devices, many of which aren't in the hands of a developer to test, especially those with limited resources.

Microsoft doesn't try to support 8-bit, 16-bit, 80x86, or even Pentium and Pentium II processors, or devices with insufficient disk or RAM. Apple no longer supports PowerPC CPUs, or 32-bit processors. Yes, you can do something with that hardware, but they no longer meet current, reasonable consumer expectations around functionality and performance. Even Debian, as an example of a mainstream Linux-based distro, has minimum requirements.

2 Likes

Well i can tell you that 32 gb for a nacked Win10 is surely not enough :smiley: I had fun recently on devices with small ssd....

If im not wrong debian dropped support for a lot of cpus 3 years ago because of kernel changes happening back then.

OK I understand now. So, I would expect less than 5 Mbps of WG throughput from a 4/32 device.

  • Perhaps in your location people don't have that much speed; but in the US and KR, some places have 100+ Mbps provisioned speed. I believe Google Fiber's first paid tier is 1000 Mbps.
  • In 2015, I actually switched BACK to OpenWrt because I realized I had a 4/32 device that only had a LAN-to-WAN throughput of ~30 Mbps. Here are WG results through an OpenWrt device on a 1000 Mbps interface:

Speed using WG:

Screenshot%20from%202019-02-27%2014-50-03

Speed (no WG):

Screenshot%20from%202019-02-27%2014-56-43

1 Like

That's what I meant @lleachii... Newer devices are not necessarily so much faster and if you consider for mesh networks at home you always need 3 or 4 devices to cover a whole house and room.

You're also right about internet speeds here. Maximum upload is still around 40 MBit/s even when you already get 250 Mbit/s down through cuper. Cable providers are still much worse. They can just offer less than 30Mb up even when they provide 1000 down already.

Yes, with fiber you can reach the physical limit of their interfaces if you find a reasonable provider for less than 100€/month if you want 1GBe sync up and down. - That's usually a bit much especially if you have absolutely no service which is able to use that bandwidth...

1 Like

I think you've missed something here, 5Ghz often works better than 2.4Ghz and you're listing ridiculous arguments in general which others also have pointed out. I just showed you one device that's cheap and is much capable of doing what you're asking for. The PSU is usually 2A tops so pretty much the same as your current devices.

This is a non-starter for me for at least three reasons:

  1. OEM firmware is generally insecure
  2. OEM consumer firmware doesn't support "advanced" networking, such as 802.11s, B.A.T.M.A.N., OLSR, GRE, VLANs, ...
  3. There are plenty of well-supported devices not in the 4/32 class that are/were moderately priced that have significantly better 5 GHz wireless performance than the either ancient or cheap SoCs and chips in the 4/32 devices (Edit: that many times don't support 802.11ac and/or MIMO, slowing down the wireless for current clients.)

Being able to route say 100Mbps and being able to enjoy 100Mbps are two different things in my opinion. I've seen 100Mbps feel so so slow when shared across a small organization (say 10 simultaneous users, like a small coffee shop) and bufferbloat leaves you with a pretty constant 800ms round trip time. Load a web page, it requires 5 or 10 DNS lookups, that takes 5 seconds, load those resources, they each require 5 different DNS sub-lookups... that takes 25 seconds... and then BLINK after 30 seconds your page shows up all at once. You didn't need to actually get more than say a megabit of data, but you waited 30 seconds for it even though if you had SQM you'd have gotten it in less than 1 second.

So, routing X speed vs SQM of X speed are two different things. And actually available 4/32 devices with say single 400MHz MIPS cores running routing/SQM/WiFi all on one core are unlikely to SQM more than 30 or 40 Mbps realistically.

To get SQM with 100Mbps you should have at least 2 cores and something in the 700MHz range or better. In reality because it's not that much more expensive than a low end device, you should get something like a WRT32X with 4 ARM cores and 1GHz+ (~$110 here in US). After all if you have 100 or 200Mbps you're paying 50 or 80 $/mo you might as well make it work well by getting enough router to be satisfied.

so, if you have low end requirements: 40Mbps or less, you will be well served by GL-iNet devices around $20 to $50, and if you have 100Mbps-250Mbps you'll be well served by WRT32X or similar NBG6817, and if you have more go x86. Nowhere in this equation is there room for a new device with 4/32 because even the cheap $20 GL-iNet device is cheaper and better

It's strictly a matter of 4/32 as legacy already existing devices that people want to continue to use.

2 Likes

I hate to state the obvious, but apparently it isn't obvious to you.

If you had one or two decent, current devices, you wouldn't need 3 or 4 ancient ones.

Not only that, but eliminating or reducing backhaul traffic, as well as unit-to-unit interference would likely further improve your wireless performance.

As much as I like the idea of using hardware as long as possible and usefull I think your approach is only for a tiny minority of users....the ones with the knowledge and equipment and lust for these mods will do it....the overwhelming majority don't care for personal (imho good) reasons.

Using older devices longer? Buy a used 8/64, 16/64 or 16/128 for little cash on ebay before their owers will throw them away because no one buys these old dogs ;- ) e.g. one NETGEAR N600 WNDR3700v2 sold for ~7,5€ P&P included another one for ~9€

...and as speaking of ressource saving: Don't throw your 4/32 device in the standard waste bin ;- )

1 Like

You got that wrong. I just installed Windows 10 Pro yesterday on a new SSD, along with Office (full package including Word, Excel, PowerPoint, Access and Outlook and some), email accounts, Norton Antivirus, Chrome, Firefox, Opera and Tor, WharsApp, and some utilities, and it's not 32 GB yet! (and hibernation is enabled).

I wish you good luck if features updates comes out and want to be downloaded and installed.

Seriously though as others have mentioned. The fundamental question is a tad one dimensional. And reframing paradigms is kinda critical here.

I for one would love these devices to remain active.... but the practicalities of hardware constraints had me repurposing and adjusting over a year ago.....

The question is as much;

"how long will these devices continue to support contemporary demands, both internally and externally......."

And from a practical level..... the best way forward is a narrow one....... what hampers advances in this domain are broad questions..... broad aims...... the scope is unrealistic and ill defined.