> With OpenWrt Two, I bet they're going to make the same mistakes as OpenWrt One: not enough memory and not upgradable, wifi not replaceable, no usable expansion slots (mini-PCI, M.2) and, of course, no (e)SATA. Another e-waste product that will be obsolete even before it's available to buy.
Those are only mistakes if you ignore the realities of what hardware is available. A highly-integrated SoC designed specifically for wireless router usage is a more cost-effective platform than a generic x86 PC. Basing OpenWRT One and Two on such hardware means work to improve support for those systems is more likely to benefit OpenWRT support for mainstream consumer networking equipment that also uses purpose-built SoCs.
OpenWRT is not yet in a position to influence the hardware design decisions made by companies like Mediatek, Qualcomm, Broadcom for their consumer WiFi product families. Those chips are still designed around what's best for the big brands that are the primary customers: Netgear, Linksys, TP-Link, etc. Adding SATA controllers to a WiFi router SoC does not benefit Netgear, et al., nor does splitting out all the radios to separate chips that could be installed onto M.2 cards (miniPCI and miniPCIe being long obsolete and bandwidth-starved). Asking for eSATA is laughably unrealistic.
A focus on the kind of modular, expandable and upgradable hardware platforms that actually currently exist (namely, PCs) is what leads to the distractions you're complaining about: "video acceleration, mesa, X, wayland, Doom, etc."
For another approach to open source networking by Linux Foundation please check DENT OS [1].
> OpenWRT is not yet in a position to influence the hardware design decisions made by companies like Mediatek, Qualcomm, Broadcom for their consumer WiFi product families.
Perhaps I'm biased, but I do believe DENT is in much better position and has more chance of influencing the white-box networking vendor than OpenWRT with regards to their design decisions.
From the website:
"As a Linux Foundation project, DENT utilizes the Linux Kernel, Switchdev, and other Linux based projects as the basis for building a new standardized network operating system without abstractions or overhead. All underlying infrastructure — including ASIC and Silicon for networking and datapath — is treated equally; while existing abstractions, APIs, drivers, low-level overhead, and other open software are simplified. DENT unites silicon vendors, ODMs, SIs, OEMs, and end users across all verticals to enable the transition to disaggregated networks."
That doesn't even appear to be attempting to address anything relevant to consumer networking. It's a purely enterprise-focused project, mostly about putting a Debian-based OS onto rackmount ethernet switches.
>That doesn't even appear to be attempting to address anything relevant to consumer networking
Don't be too pessimistic, anything that can contribute to the disaggregation of networking technology will be good for the consumer markets since the companies producing these networking SoC namely Qualcomm, Mediatek, etc should care about their downstream revenues. By adopting the open eco-system champion by SONiC and DENT for networking technology disaggregation, it will definitely spill over to the consumer networking as well since these are the very same companies that design and manufacture the networking pervasive consumer networking SoC.
There are too many and numerous examples of Linux enterprise features that are spilling over to consumer Linux, and the latest is the real-time Linux kernel extensions although it did take like forever (i.e 20 years) to be adopted by the main line Linux kernel.
> Those are only mistakes if you ignore the realities of what hardware is available.
If you think the focus should be on what hardware is available, then why make a OpenWrt Two instead of buying the existing hardware?
> A highly-integrated SoC designed specifically for wireless router usage is a more cost-effective platform than a generic x86 PC.
This is exactly why OpenWrt One and probably Two too is just e-waste - because those cheap integrated hardware platforms are e-waste to begin with. They are indeed cost-effective, but only for a brief moment in time.
Wifi drivers are one of the most problematic part of linux kernel. Also, wifi standards are still changing very fast. Non-replaceable wifi is one of the things that's going to kill these boards.
OpenWrt One can only be used as a wireless router. There's no storage, no expansion slots, not even USB3. It can't be repurposed, can't be upgraded, can't even be used as an ordinary ethernet router because there's no switch. In less than a year, OpenWrt Two makes it obsolete. OpenWrt Two won't be any different, so why make it at all? What will that improve? There's tons of boards better than that (BananaPi R-series, GLinet routers).
So that's basically my argument.
(I didn't complain about the CPU. The CPU is probably the most future-proof component in there. Using x86 CPU is probably the worst design decision they could make.)
> Adding SATA controllers to a WiFi router SoC does not benefit Netgear, et al.
So what? That's what PCIe is for (and expansion slots).
Turris Omnia is almost 10 YEARS old and it's still being sold, and at outrageous prices, mind you, even second-hand. It's CPU is obsolete, it's miniPCIe slots are obsolete, it's ethernet is (almost) obsolete, it's memory is barely sufficient, and yet it's still usable as a home router, personal web server, file server, NAS, torrent client, remote download manager, TOR node, proxy, etc. etc. after 9 years! Why is that? Because Wifi was replaceable and it had 2GB RAM at a time when most routers only had 32MB.
Beat that, OpenWrt One&Two!
> A focus on the kind of modular, expandable and upgradable hardware platforms that actually currently exist (namely, PCs) is what leads to the distractions you're complaining about: "video acceleration, mesa, X, wayland, Doom, etc."
No, that's not it. A mini-PCIe / M.2 slot won't fit a GPU (well, it could, but...). A SO-DIMM slot instead of soldered memory also won't change anything. Meanwhile, GPUs are already there in most SoCs supported by OpenWrt, not just x86: Rockchip, Mediatek, Broadcom/RaspberryPi, you name it.
If devs were interested in GPU support, they could have contributed to Buildroot instead. Why did they add it to OpenWrt instead of Buildroot? Probably because it was easy: they're the main devs of OpenWrt with commit privileges. I'm not saying that they abused those privileges. I'm saying that their interest doesn't seem aligned with what OpenWrt is: an OS for routers.
> modular, expandable and upgradable hardware platforms that actually currently exist (namely, PCs)
And this is the second problem with OpenWrt One/Two. If an OpenWrt dev wanted to have an OpenWrt NAS, or an OpenWrt server of some kind, anythining other than (just) a router, only PCs fit the requirements. Not even Rockchip/BananaPi SBCs. Of course that dev is going to contribute with support for PCs in OpenWrt.
Am I understanding you correctly: your complaint about OpenWRT software is that they aren't focused exclusively enough on being a router, and your complaint about OpenWRT hardware is that they are focused exclusively on being a router?
- There's no hardware for router+NAS functionality, other than repurposing x86 PCs.
- OpenWrt software is not good enough at routing, could be a little better as NAS, but it seems now focused on being usable as a general-purpose desktop OS.
Those are only mistakes if you ignore the realities of what hardware is available. A highly-integrated SoC designed specifically for wireless router usage is a more cost-effective platform than a generic x86 PC. Basing OpenWRT One and Two on such hardware means work to improve support for those systems is more likely to benefit OpenWRT support for mainstream consumer networking equipment that also uses purpose-built SoCs.
OpenWRT is not yet in a position to influence the hardware design decisions made by companies like Mediatek, Qualcomm, Broadcom for their consumer WiFi product families. Those chips are still designed around what's best for the big brands that are the primary customers: Netgear, Linksys, TP-Link, etc. Adding SATA controllers to a WiFi router SoC does not benefit Netgear, et al., nor does splitting out all the radios to separate chips that could be installed onto M.2 cards (miniPCI and miniPCIe being long obsolete and bandwidth-starved). Asking for eSATA is laughably unrealistic.
A focus on the kind of modular, expandable and upgradable hardware platforms that actually currently exist (namely, PCs) is what leads to the distractions you're complaining about: "video acceleration, mesa, X, wayland, Doom, etc."