Imagine if linksys would produce a platform specifically for the open-source community to use.  What features would you want to include.   For me, the following would be ideal;

  The serial port(s) being wired out to the outside of the case on an external ( rj45 ) connector
  Two USB Ports
  easy access to the JTAG port, either via an external conenctor or at the least a header plug
  16MB of flash, 32MB RAM or should that be 64MB
  A PCMCIA slot

What else could be useful.   There is opportunity to take this back to Linksys( cisco ) and give them some pointers as what to do...   They are interested.

My #1 priority would be a PCI slot... Then you could plug almost anything in (even a PCI->PCMCIA adapter, or a dual miniPCI adapter, or...)

I know it would take some time as we would need to port to a device I would like to see, and it would be more expensive then a wrt, but if we are speaking of the 'perfect' one, here's the list:

'The good' (eg. what is needed imho):
Intel IXP42x CPU
RedBoot or u-boot bootloader
Atheros wireless in mini-PCI slot, not soldered.
External serial, JTAG headers onboard
at least 8MB flash and 32MB ram

'the bad' (eg. what would be kickass):
So-DIMM memory socket instead of soldered memory (with 32MB as default, so enough for most cases, but if people want to extend it, it's possible)
Gbit switch (sure, it would connect to the CPU with only 100Mbits/s, but the LAN would be on Gbit)
USB2 ports
POE support

'and the ugly' (eg. what to avoid):
Broadcom wireless cards or anything with binary-only drivers

If we take a look business-wise:

start with an opensource project from the beginning
sell only the hardware (eg. let the community deliver firmware for the device)
maybe make two versions, 'the good' and 'the bad'

if You take a look up there, the WRV54G would need only small changes to be the perfect one. For 'the good' they would only need to change the bootloader, the wireless card and add an external UART. Oh, and don't add the vpn features by default.

I think the most important thing a company could offer would be full and complete documentation for the hardware. If you look at the source of drivers that come out of places like Broadcom and TI, you'll see lots of cruft needed to let the driver work on everything from Linux down through VxWorks through ThreadX through to crappy in-house OSes that barely provide the primitives their driver uses (look at the AR7's avalanche_cpmac Ethernet driver for an example).

With good documentation, it becomes possible for the community to write its own driver directly, without trying to work out what they need to do to the hardware from the code. Plus, it provides us with an easy way to look for driver/hardware interaction issues.

today is in the market many of such boards/platforms (worp, routerboard, etc.), but price is $150-$300 without case and power supply. Devices, what is produced by high volumes for the dumbusers is always better for the open projects  than perfect and expensive ones.

It's a matter of how expensive and how big the volumes are. Since linksys has a <10,000 estimate on expected sales of the WRT54GL / month. It is also worth noting the success the people working on pchdtv.com had.

I could see design cost adding to the price of low volume products. However, broadcom has some reference designs (hopefully meaning they can be copied for free). There are some PCB companies which turn specifications into boards. The cost of having a few prototype boards made should be a few hundred at most. I don't know how much soldering the stuff on them would cost, but someone can solder them if the cost is prohibitive. Once a functional circuit board has been made, I don't know how much a contract manufacturer will charge to make a few thousand of those routers. I think I saw somewhere that the BOM of a comprable router made from the broadcom chips is $30-40. I think the price of the broadcom chip was ~$10.

I would pay ~$100 for such a router w/ a normal PCI slot, or better yet, a riser w/ 4 PCI slots, boosted DRAM and flash.

(Last edited by linbe on 16 Jan 2006, 01:08)

In the meantime, it's exactly vice versa. Atheros are binary-only as well, BTW (at least the HAL crap).

More flash+RAM, serial port, JTAG. PoE wired on port 1. With a flexible bootloader.

USB would be a nice extra, but I'd be happy enough with just an improved version of the old GS v1.0.

(Last edited by vincentfox on 17 Jan 2006, 05:54)

For me, ASUS WL-500g Deluxe is close to being perfect.

The only thing I'm missing is a slot for more RAM.

Lack of ported apps is also a problem, but I think it can be solved by the community once there is a need for a given app.

I intend to use OpenWRT + these tiny routers as a replacement for real PC servers in small offices.

Just imagine that you can have an all-in-one router, print server and a domain controller, closed in a small, fan-less and disk-less box!

Linksys could just take the existing NSLU2 design, add a miniPCI socket for expansion, put a PHY on the second unused NPE to get two distinct Ethernet interfaces. They could then attach a 5 port VLAN switch to one of these Ethernet interfaces. Putting JTAG pins on the board and providing a serial console port as standard would round out the box quite well.

If they could retail a box like that for about US$150, they'd be onto a winner. The existing NSLU2 community already proved that they can do some pretty neat stuff.

In my books the most important features are:

Ethernet (1 or 2 interfaces)
USB2 (1-4 host ports, peripheral port for bonus points)
Serial console with RS-232 levels (secondary serial for use by applications would be good)
8-32 MB flash
32-64MB RAM
802.11g (prefered on miniPCI)
VLAN switch
JTAG
Specs and databooks for all hardware

The last item is very important if the full potential of the device is to be harnessed by the OS community.

There was some talk on the forum about Senao producing such a board based on the Atheros SoC.  Someone mentioned sometime in the next couple of months as the launch.  It is supposed to be Linux friendly, so hopefully we can get OpenWrt onto it quite quickly, building on all the existing work.

i just want an openwrt AP that doesnt have a sucky radio.