Topic: Anyone interested to design own OpenWRT Hardware ?

Hello !

We are a small german development company and are currently designing our own Wireless Router based
on the RALINK 3662f APsoc (500 MHZ MIPS). The result will be someting similar to the ASUS RT-N56U - except
with 1 GB FLASH !

So I think this would be perfect for OpenWRT. On the one hand we will use the hardware for our own products
(Firewall stuff) - but on the other hand we will try to sell these Systems rather cheap for OpenWRT - Users. (< 100€).

Currently wee need some Input and Help to get everything "to fly":

- Suggestions from the OpenWRT Developers (What DSL Controller to use, WHat switch to use, etc.)
- Help from the Developers: Get this ralink evaluation board (reference design) running with openwrt.

Anyone interested to participate ?

Frank

Re: Anyone interested to design own OpenWRT Hardware ?

fmenne wrote:

We are a small german development company and are currently designing our own Wireless Router based on the RALINK 3662f APsoc (500 MHZ MIPS).

Hmm, which MIPS-CPU is this?  http://wiki.openwrt.org/doc/hardware/cpu#mips Why?


Which SoC will this be? Will this device have wireless? According to  http://en.wikipedia.org/wiki/Comparison_of_open_source_wireless_drivers there are FOSS drivers for Ralink available. But not for all Ralinks... if you want to sign NDAs and use their proprietary stuff, you should ask for help at DD-WRT. I think they cooperate on this level with Broadcom. (This is one of the reasons, why the ASUS-devices with Broadcom-SoC are supported by DD-Wrt but not (yet) by OpenWrt. I don't think you are going to find help here with that.


The result will be something similar to the ASUS RT-N56U - except with 1 GB FLASH !

WTF? Why??? Only stupid people are impressed by that! Give us more RAM. Give us more RAM.

There are already devices with 128MiB. For 100 Euros. If you are smart, you will offer 256MiB (in german this is called "Alleinstellungsmerkmal", and this would be a good one). Now, on a usual router, you do not need 256MiB, but if you use OpenWrt, you WANT 256MiB and more! And if you really want to impress people, you will offer not less then 512MiB of RAM. Yeah, for 100 Bucks.

Regarding the Flash: You are going to want enough, that you can install a good OpenWrt System on it. I could imagine, that 256MiB are more then enough! If want to install something different, like say a Debian, 256 will not be sufficient. But Flash is constricted. Because of the wear: no extensive logging, no journaling FS, NO TORRENT!, etc. And the flash is often _very slow_. I don't mean the expensive SSDs, but cheap sticks, that offer around 5MB/s write speed. The soldered one will not be much quicker. The connection is another question.

The flash is only good, to install the bootloader on (let's say 1MiB is sufficient) and some programs you read in case of emergency and want to have handy even without a hard disc. I could imagine this basic stuff fits just fine on 32MiB of flash. I would use a hard disc for anything else, and if used a harddisc, I would install the whole system on it, for painless upgrades.

You could read this:
http://wiki.openwrt.org/doc/toh/buyerguide
http://wiki.openwrt.org/toh/seagate/dockstar/install.location

What you basically want to offer, is a mini-server. Well, at least that is what I am interested in, and there is only shit on the market. All that ARM-stuff come with shitty graphics solution, with no Linux drivers at all, at least not available ones, closed source drivers without specs (Tegra...), etc. The old stuff, like Feroceon, has only 16bit RAM Bus.

I would like to try something like the dockstar, but with MIPS (I am most curious about this, so far you see a lot of ARM everywhere), something like CESA but for 2048 bit and more RAM, an integrated GBit Switch and maybe wireless. Personally, I could live without the wireless ;-) Built-in DSL-modem could be nice, my stand-alone DSL-modem consumes 14W :-O !

So I think this would be perfect for OpenWrt. On the one hand we will use the hardware for our own products
(Firewall stuff) - but on the other hand we will try to sell these Systems rather cheap for OpenWRT - Users. (< 100€).

What will be different from all the hardware already available??? 1GiB of flash? I don't need 1 GiB of flash. If wanted that, I'd buy a cheap usb stick with 4GiB of flash. Oh, if you include a USB port, make sure it is able to power a 2,5" harddisc. This would be your second unique selling point. Not an important one, though.

Currently wee need some Input and Help to get everything "to fly":

- Suggestions from the OpenWRT Developers (What DSL Controller to use, WHat switch to use, etc.)

I am not a developer, so I cannot help you. Ask AVN, the Fritzbox dudes, regarding the DSL-stuff and DD-WRT if you want to use non-FOSS drivers.

- Help from the Developers: Get this ralink evaluation board (reference design) running with openwrt.

Anyone interested to participate ?

Frank

Sorry, I cannot help you with that either. But if you really care about what hardware we would buy => give us more RAM... ;-)

This  http://www.wehavemorefun.de/fritzbox/index.php/FRITZ!Box_Fon_WLAN_7390 costs about 200 Euros (hopelessly overpriced if you ask me). Has a build-in VDSL-Modem, and I am pretty sure the drivers for the modem are ClosedSource. I have no idea how much they modified the kernel and other stuff, but I understand that they invested some money in the development of their software.

You could ask them, or have a look at the Freetz-Software. This should include xDSL-drivers as well. I guess again ClosedSource. I haven't heard of FOSS drivers for any DSL-modem, but I haven't searched.

3 (edited by NetworkPro 2011-02-22 00:46:45)

Re: Anyone interested to design own OpenWRT Hardware ?

Think of WISPs too.

Think of software too : https://forum.openwrt.org/viewtopic.php?id=28760

P.S. Think of VoIP too please.

Re: Anyone interested to design own OpenWRT Hardware ?

Sounds very interesting!
I'm just in touch with some fellows (related to TP-L....). But to get 1/10 of a container sad To less probably..

Means, we need 100-200/year or so, pretty few.. Don't care about HW-details as long as it has a 32Bit CPU, 8+ Flash and 32(+) RAM and  - most important - at least one USB! (Ok, Eth+Wlan would be helpful still wink)

But also: really cheap! (remember: an Alix with a LX800, 256MB RAM (not 16..32..) is about 80 EUR + CF in GB, not in 2-8 MB-range.. And thats "old stuff", plenty Platforms below 100 EUR are there giving GHz and GB's) [honestly, I wouldn't fiddle around with a Mips 8/32 MB in private, just spend 40 EUR more and install a stock Debian.. but in hundreds it matters..]

If we could get pretty below that (max 50 EUR net..): tell/call me please (it should also be possible to find me by my realname in my profile wink)

Makki

Re: Anyone interested to design own OpenWRT Hardware ?

fmenne wrote:

but on the other hand we will try to sell these Systems rather cheap for OpenWRT - Users. (< 100€).

good luck

All the waiting and then the CPU ventilator brrrrrr all the time.

6 (edited by KanjiMonster 2011-02-22 15:04:24)

Re: Anyone interested to design own OpenWRT Hardware ?

Orca wrote:
fmenne wrote:

We are a small german development company and are currently designing our own Wireless Router based on the RALINK 3662f APsoc (500 MHZ MIPS).

Hmm, which MIPS-CPU is this?  http://wiki.openwrt.org/doc/hardware/cpu#mips Why?

The rt3662/3883s have a MIPS 74K.

Orca wrote:

The result will be something similar to the ASUS RT-N56U - except with 1 GB FLASH !

WTF? Why??? Only stupid people are impressed by that! Give us more RAM. Give us more RAM.

There are already devices with 128MiB. For 100 Euros. If you are smart, you will offer 256MiB (in german this is called "Alleinstellungsmerkmal", and this would be a good one). Now, on a usual router, you do not need 256MiB, but if you use OpenWrt, you WANT 256MiB and more! And if you really want to impress people, you will offer not less then 512MiB of RAM. Yeah, for 100 Bucks.

Agreed with more RAM. But the Soc supports only 256MiB, so that's the limit.

Also please, please use NOR Flash instead of NAND. NAND is a PITA, while NOR is much more reliable.

fmenne wrote:

Currently wee need some Input and Help to get everything "to fly":

- Suggestions from the OpenWRT Developers (What DSL Controller to use, WHat switch to use, etc.)

My suggestion for DSL would be Lantiq (formerly Infineon), they have open source DSL drivers which are pretty well supported in Openwrt. I suggest you get in touch with John Crispin for further info, he's also the maintainer for the Lantiq platform in OpenWrt.

Regarding the switch, I would suggest Atheros since I rather like their stuff and they were quite actively supporting open source, but unfortunately I can't say if their stance changed since they got bought by Qualcomm. It does not seem like it changed, looking at the current state of Ath9k. Using a gigabit capable switch should be a given.
Realtek also produces switch chips, but I can't comment on their quality.

fmenne wrote:

- Help from the Developers: Get this ralink evaluation board (reference design) running with openwrt.

Okay, so it seems the SoC is already set, so no 3T3R with the rt3883 anymore wink. You might want to get in touch with Juhosg, he seems to be the maintainer of OpenWrt's ralink support.

Also, while it certainly would ramp up the price, having a Mini PCIe slot would be a huge plus in my eyes. This would allow anyone to use their favourite wifi chip with it. Also Ralink wifi support in Linux is unfortunately of rather questionable quality/stability (at least the open source variant), so having Atheros wifi would be quite welcome, too.

fmenne wrote:

Anyone interested to participate ?
Frank

If you choose to use an Atheros switch, I can probably help wink.

KM

P.S: Also please don't make it look anything like the RT-N56U. I think its quite ugly. Also avoid "Klavierlack-Optik", I hate it with a passion wink. Oh, while were on the topic on external hardware features: Detachable antennas, always a plus.

Okay, since now I likely made it at least twice as costly, I should probably stop here. wink

Re: Anyone interested to design own OpenWRT Hardware ?

Hi, Friends

as I am new in such kinds of community I must first adjust a little bit to the "rough" wording.
Understand me right: We do not want to get rich on OpenWRT users - we are making our profits with our own products.

What I have learned from the discussion so far: All of you have good experience in this technical stuff and we can profit from each other.

It seems to be clear that we should increase RAM. Should not cost that much.
We will try to include NOR Flash and maybe would downsize flash size on the other hand to keep costs low.
RT3883:Pinout is identical. Both are BGA - take a Hair-blower and a baking oven... ;-)
Regarding Switch: would the RTL8367 also be acceptable ? What would be the equivalent from Atheros ?
Mini - PCI: Maybe we could prepare this on the Mainboard and you only have to install the costy connector.

Frank

Re: Anyone interested to design own OpenWRT Hardware ?

I would prefer not to have to solder on the board, if that's what you mean by adding the miniPCI/miniPCI-e connector ourselves? smile

It is considered a sales point for a product to be easy to use.

Re: Anyone interested to design own OpenWRT Hardware ?

Hi, Friends

certainly you are right - The connector itself seems to cost 40 Cent per piece from our local Electronic Distributor.
So I will add Mini-PCI to the Whishlist. ;-)

Frank

Re: Anyone interested to design own OpenWRT Hardware ?

fmenne wrote:

Hi, Friends

certainly you are right - The connector itself seems to cost 40 Cent per piece from our local Electronic Distributor.
So I will add Mini-PCI to the Whishlist. ;-)

Frank

example of an existing hardware http://wiki.openwrt.org/toh/d-link/dir-825
You'd need to go below 50Euro

All the waiting and then the CPU ventilator brrrrrr all the time.

Re: Anyone interested to design own OpenWRT Hardware ?

Hi, Friends

I checked several price engines (Amazon, etc.): Dlink is at approx. 100€. With 8 MB flash.

Re: Anyone interested to design own OpenWRT Hardware ?

fmenne wrote:

Hi, Friends

I checked several price engines (Amazon, etc.): Dlink is at approx. 100€. With 8 MB flash.

93 on idealo.de
and 8mb flash is not a problem really

All the waiting and then the CPU ventilator brrrrrr all the time.

Re: Anyone interested to design own OpenWRT Hardware ?

Hmmm, AVM used AR7 and UR8 which are/were both Texas Instruments SoCs:   http://www.linux-mips.org/wiki/AR7#AR7_Ownership
They soled it again and again, and today this is property of Lantiq.

The last device from AVM is not based on UR8 any more, but on  Ikanos Fusiv Vx180.  http://www.ikanos.com/products/broadband-dsl/
You'll find some links here: http://wiki.openwrt.org/toh/avm/fritz.box.wlan.7390

And this device is aimed to handle VoIP and IPTV as well as POIS (Plain old internet service) ;-)
Now AVM is said to offer good products, but I take offense in their pricing. I can get a PC for that money.
But I am even more offended by companies such as Alice and their IAD "products", here is one:

http://www.wehavemorefun.de/fritzbox/index.php/Alice_IAD_7570

1. It is and remains their property, so you cannot (well, or shall not) flash Freetz or OpenWrt on it

2. They do not sell it (and they do not distribute it either because it remains their property and is (considered to be) part of their net, like the DSL-AC, so they do not need to follow GPL. So far, I do not know, if they altered to much or anything, but we have to look into the future.

3. AFAIK they do not give you the data needed to use their connection with some other device, like a real FritzBox. VDSL should be no problem, but user complain it is not easy to get VoIP working or not possible at all.

Question is not what we want, but what you're company is planing to do. If you want to "bless" us with more products like the IAD shit, good luck in finding help here. If you want to give AVM some concurrence, I hope for lower pricing ;-) I don't need new hardware, but if I'd buy something, it would have to replace: Splitter, NTBA, Modem, Router (I want OpenWrt on it), etc. and consume less then 10W max. If I could operate a 2,5" harddisc without a power pack, that would be cool. I want enough RAM, so there will be no problem with Asteriks (or what ever PBX-software you/I put on it for VoIP), torrent with enough RAMcache and all the other stuff I neeeed to have running 24/7. (Maybe OpenVPN-Server to make WLAN a bit more secure. Just give me the RAM, let me fill it).

This Fritz 7390 seems quite remarkable, but it is far to expensive and does not nearly offer the functionality of OpenWrt (even Freetz does not).

Re: Anyone interested to design own OpenWRT Hardware ?

fmenne wrote:

Hi, Friends

certainly you are right - The connector itself seems to cost 40 Cent per piece from our local Electronic Distributor.
So I will add Mini-PCI to the Whishlist. ;-)

Frank

Cool. And make that mini-PCI express; mini-PCI is quite outdated wink - possibly with a SIM card slot, so one can put a 3G modem in there - normally the modems don't come with a SIM card reader.

fmenne wrote:

Regarding Switch: would the RTL8367 also be acceptable ? What would be the equivalent from Atheros ?

I assume you mean the RTL8367M (since you have two RGMII interfaces, and you need to connect them somehow to the switch ;-). Btw, which the RT-N56U also has.

I can't really comment about the quality of the Realtek switches, but have seen them more often pop up with problems on the OpenWrt tracker. This might be corellated with Realtek switches being the most common ones on non broadcom boards, so it probably doesn't mean anything.

Featurewise the AR8327 would be most equivalent chip.

They share most features like 4k VLAN table, 4 hardware queues, ingress/egress bandwith limiting, QinQ, etc; the Atheros one has a 50% bigger ACL table and supports PPPoE header removal, while the Realtek one seems to support LLDP in hardware and has a bigger MAC lookup table (which all probably doesn't matter anyway in your typical consumer applications).

USB was already mentioned, so only a +1 from me here. Also prepopulated Serial/JTAG headers! wink.

KM

Re: Anyone interested to design own OpenWRT Hardware ?

If possible - 2-3x miniPCI and 2-3 miniPCI-e connectors smile

miniPCI is popular in WISPland.

Re: Anyone interested to design own OpenWRT Hardware ?

Hi, Friends

mini-PCI- express - Yes, Sir. ;-)
2 USB ports should be no problem without additional hardware.
EJTAG 14 Pin Header and Serial Header (3.3V level) are also my choice.

Does someone have an idea which PC JTAG interface to use for initial programming ?
Thought of http://www.usbjtag.com/ .... but I am not sure what is suitable for RAMIPS architecture....

F.

17 (edited by Insecure 2011-02-23 12:27:26)

Re: Anyone interested to design own OpenWRT Hardware ?

fmenne wrote:

Does someone have an idea which PC JTAG interface to use for initial programming ?
Thought of http://www.usbjtag.com/ .... but I am not sure what is suitable for RAMIPS architecture....

Quote from the ddwrt wiki
USB Jtag is expensive, and doesn't appear to work consistently well. The best option is to get a ExpressCard Parallel port adapter.

So i would perhaps go for this one http://www.diygadget.com/jtag-cables/universal-jtag-adapter-for-routers-modem-fta-and-more.html i don't know if it's cheaper though.

Re: Anyone interested to design own OpenWRT Hardware ?

The USB jtag is fast and has software. What software does the diy jtag have?

If they are going to provision a lot of boards they would need a fast JTAG or better programming solution.

19 (edited by Orca 2011-02-23 12:41:11)

Re: Anyone interested to design own OpenWRT Hardware ?

fmenne wrote:

2 USB ports should be no problem without additional hardware.

Two USB ports are nice, but nothing special. If the USB could provide enough power/electrical current/energy, so one could attach a 2,5" Hard disc to it, without a dedicated power supply, that would be nice.

Also, some people ask about JumboFrame support (that is, if you'd have 1000BASE-T Ethernet). Some switches come with that, but most (or all) PHYs do not support JumboFrames.

Now it does not make sense to use Jumbo-Frame and use internet at the same time, since this only leads to Fragmentation and that again to more CPU load. Or can you? The software (on my PC) is not clever enough to distinguish between packets into the internet and internal ones. It's one connection, one MTU. Sad.

Well, some scenarios could still make good use of JumboFrames especially because (I think) that the many interrupts and not the limited CPU power is responsible for lower then USB samba/NFS/ftp throughput.

Take a look at the tags:  http://wiki.openwrt.org/meta/tags

Re: Anyone interested to design own OpenWRT Hardware ?

JumboFrames is used in Data Centers and sometimes by ISPs. If this board is going to be "so good" so they would use it instead of some more serious hardware...

It would need a rack-mountable case or to be compatible with the ones already available for MikroTik RouterBOARDs.

Re: Anyone interested to design own OpenWRT Hardware ?

According to the Ralink drivers, the RT3662/3883 supports a maximum frame length of 4096, so no 9K jumbo frames with Ralink (but 4K is an improvement).

Re: Anyone interested to design own OpenWRT Hardware ?

NetworkPro wrote:

JumboFrames is used in Data Centers and sometimes by ISPs. If this board is going to be "so good" so they would use it instead of some more serious hardware...
It would need a rack-mountable case or to be compatible with the ones already available for MikroTik RouterBOARDs.

Most NICs are capable of some sort of JF (and quite often not more than 4K) and you never know when you could need them. The PC has more then enough power, but the router/server does not. Because of this, such a device would benefit the most of JF! Problem is, you just cannot have it all at once, so the use of JF remains very limited. An alternative would be a technique they use with 10GBit NICs instead of JumboFrames. That not every packet pulls an IRQ, but only every 100.000 or so. But would netfilter work with this? Ahh, plenty of ideas.

My point was the "unique feature". I don't see, how they are going to compete on this market without a couple of unique selling points. Howto compete with TP-Link? Chinese only, so no high salaries in production and in developing, and they use hardware which has FOSS drivers. Not only do they save on money by that, but actually we also prefer FOSS drivers. They do lack a 5GHz capable router, maybe they will bring out something with enough RAM to exhaust the possibilities.  And again, at a fair price. That is, if this new company searched for a device to sell to the public in the first place. And I don't think they are ;-)

I do not expect that the ideas here will be fullfilled, they probably need some thing adapted to a certain special scenario but I am still happy to offer some input. You never know, maybe some good product, available to the public to buy, will nevertheless  come out of this. I am most curious.

Re: Anyone interested to design own OpenWRT Hardware ?

Orca wrote:

My point was the "unique feature". I don't see, how they are going to compete on this market without a couple of unique selling points. Howto compete with TP-Link?

yep TL-WR1043ND is about 50Euro.

All the waiting and then the CPU ventilator brrrrrr all the time.

24 (edited by NetworkPro 2011-02-25 15:07:33)

Re: Anyone interested to design own OpenWRT Hardware ?

It would maybe compete with products like this http://www.geek.com/articles/gadgets/dreamplug-puts-a-1-2ghz-arm-pc-in-a-power-outlet-2011022/

dual gigabit, 1.2GHz ...

P.S. Thanks for the consumer advice from the following post.

Re: Anyone interested to design own OpenWRT Hardware ?

NetworkPro wrote:

It would maybe compete with products like this http://www.geek.com/articles/gadgets/dreamplug-puts-a-1-2ghz-arm-pc-in-a-power-outlet-2011022/

He asked for a Switch, so I don't really think so. Besides: 150 Bucks???? For what?  For 170 you get a http://pandaboard.org/ or even better a fully matured PC. A used notebook with keyboard and monitor. I bought  the dockstar for 25 Bucks, and it's worth that much. But a hundred and fitty? I don't think so. Yes, there are things like the "emotional selling proposition", but I am too much of a geek to fall for that. This hardware, this equipment, is not worth 150 Bucks. How did they calculate the cost?
  - "The guru plug is available since 2005 and costs 100 Bucks, we add the audio and the JTAG, that makes it 120. Then the cool panting, hmm, let's make it and hundred and fitty."
Bravo! Genius.

And if you wanna complain about the high costs of development... Instead of squirting around with the sheeva plug and guru plug and the guru plug plus plus and now this stuff, somebody should have designed a useful device the first time and sold it at a decent price. The amount of sold units is key. Some capable developers without interference from other departments, could do the trick.

Intel and AMD put sooo much more silicium in their products then ARM and MIPS, and still they manage prices, you could call decent. Now compare that to the complexity of ARM and MIPS CPU designs, consider that we use SoCs, and explain the high prises. WTF? And if you want to sell more units, make a better and more generally deployable product!!!
You behave like the guys who sell Laptops with double 3GHz CPU and shit and with 1GiB of RAM. 2GiB cost 70 bucks extra. 4GiB are only possible with their "deep shit high end" product... Ridiculous.

And yes, I know about   http://en.wikipedia.org/wiki/Kaizen   and it can go it both directions. Instead of merely "improving" your product over time, you end up selling bananas. Immature shit. Hey, like in the software-sector... People who bought the first PS3, now have a device in their living room that consumes about 200W to simply watch a BluRay. Great job. For playing, that is ok, but 200W for watching a BluRay??? Ridiculous. And the prices for electrical current are going to RISE, my friends. No invasion is going to avoid that. Look at the plans of the german government for the future. This is going is cost plenty. Someone will have to pay for all that sudden "innovation" in the energy sector.


NetworkPro wrote:

dual gigabit, 1.2GHz ...

LOL,  dual "Gillion" with 16bit-memory bus connection and the floating point performance from 1989. Impressive. For a 150 Bucks. Well, at least it has CESA. How is the CESA performance? How the support? Not the problem of the company who sells the product ? Hmmmm.... well, it can "do sound".

BTW, how do they manage to support ARMv5 and ARMv7 and ARMv9 at the same time? I guess enough people to compile and support software for all of it. Or not?

Not your problem? Well, then don't complain you don't sell enough units and cannot offer us decent prices.

Oh, after you come up with a good design, that would perform for the next couple of years, you could (finally) concentrate your energy on the software support. Start with the drivers. And if you do not have the manpower to support your (stable and performent) closed source drivers for a couple of kernel versions, maybe you should "risk it" and go GPL.


Here is some benchmarking for kirkwood and TI OMAP and Tegra (sadly no MIPS):
http://dev.gentoo.org/~armin76/arm/memcpy-neon_result.txt
http://dev.gentoo.org/~armin76/arm/nwbuildtimes.xml
Then some MIPSs CPUs but for OpenSSL:    http://wiki.openwrt.org/inbox/benchmark.openssl

Now before you cheer for the Tegra, check their drivers... And will they support wayland? No? Ohhh, that's sad. Now check driver's for the PowerVR, hmm, yeah. Will there be intel or AMD graphics with MIPS/ARM-CPU? Sad again.

But the most cool stuff is the actual software anyway. And for umurmur, asterisk, torrent, dc-hub, webserver you need RAAAAAM and not computing power. Some howtos to configure them fast are also nice to have ;-)